*509Chief Justice Roberts
delivered the opinion of the Court.
People make mistakes. Even administrators of ERISA plans. That should come as no surprise, given that the Employee Retirement Income Security Act of 1974 is “an enormously complex and detailed statute,” Mertens v. Hewitt Associates, 508 U. S. 248, 262 (1998), and the plans that administrators must construe can be lengthy and complicated. (The one at issue here runs to 81 pages, with 139 sections.) We held in Firestone Tire & Rubber Co. v. Bruch, 489 U. S. 101 (1989), that an ERISA plan administrator with discretionary authority to interpret a plan is entitled to deference in exercising that discretion. The question here is whether a single honest mistake in plan interpretation justifies stripping the administrator of that deference for subsequent related interpretations of the plan. We hold that it does not.
I
As in many ERISA matters, the facts of this case are exceedingly complicated. Fortunately, most of the factual details are unnecessary to the legal issues before us, so we cover them only in broad strokes. This case concerns Xerox Corporation’s pension plan, which is covered by ERISA, 88 Stat. 829, as amended, 29 U. S. C. § 1001 et seq. Petitioners are the plan itself (hereinafter Plan), and the Plan’s current *510and former administrators (hereinafter Plan Administrator). See § 1002(16)(A)(i); App. 32a. Respondents are Xerox employees who left the company in the 1980’s, received lump-sum distributions of retirement benefits they had earned up to that point, and were later rehired. See 328 F. Supp. 2d 420, 424 (WDNY 2004); Brief for Respondents 9-10. The dispute giving rise to this case concerns how to account for respondents’ past distributions when calculating their current benefits — that is, how to avoid paying respondents the same benefits twice.
The Plan Administrator initially interpreted the Plan to call for an approach that has come to be known as the “phantom account” method. 328 F. Supp. 2d, at 424. Essentially, that method calculated the hypothetical growth that respondents’ past distributions would have experienced if the money had remained in Xerox’s investment funds, and reduced respondents’ present benefits accordingly. See id., at 426-428; App. to Pet. for Cert. 146a. After the Plan Administrator denied respondents’ administrative challenges to that method, respondents filed suit in federal court under ERISA, 29 U. S. C. § 1132(a)(1)(B). See 328 F. Supp. 2d, at 428-429. The District Court granted summary judgment for the Plan, applying a deferential standard of review to the Plan Administrator’s interpretation. See id., at 430-431, 439. The Second Circuit vacated and remanded, holding that the Plan Administrator’s interpretation was unreasonable and that respondents had not been adequately notified that the phantom account method would be used to calculate their benefits. See 433 F. 3d 254, 257, 265-269 (2006).
The phantom account method having been exorcised from the Plan, the District Court on remand considered other approaches for adjusting respondents’ present benefits in light of their past distributions. See 472 F. Supp. 2d 452,456-458 (WDNY 2007). The Plan Administrator submitted an affidavit proposing an approach that, like the phantom account method, accounted for the time value of the money that re*511spondents had previously received. But unlike the phantom account method, the Plan Administrator’s new approach did not calculate the present value of a past distribution based on events that occurred after the distribution was made. Instead, the new approach used an interest rate that was fixed at the time of the distribution, thereby calculating the current value of the distribution based on information that was known at the time of the distribution. See App. to Pet. for Cert. 147a-153a. Petitioners argued that the District Court should apply a deferential standard of review to this approach, and accept it as a reasonable interpretation of the Plan. See Defendants’ Pre-Hearing Brief Addressed to Remedies in No. OO-CV-6311 (WDNY), pp. 7-8; Defendants’ Pre-Hearing Reply Brief Addressing Remedies in No. OO-CV-6311 (WDNY), p. 2.
The District Court did not apply a deferential standard of review. Nor did it accept the Plan Administrator’s interpretation. Instead, after finding the Plan to be ambiguous, the District Court adopted an approach proposed by respondents that did not account for the time value of money. Under that approach, respondents’ present benefits were reduced only by the nominal amount of their past distributions— thereby treating a dollar distributed to respondents in the 1980’s as equal in value to a dollar distributed today. See 472 P. Supp. 2d, at 457-458. The Second Circuit affirmed in relevant part, holding that the District Court was correct not to apply a deferential standard on remand, and that the District Court’s decision on the merits was not an abuse of discretion. See 535 F. 3d 111, 119 (2008).
Petitioners asked us to grant certiorari on two questions: (1) whether the District Court owed deference to the Plan Administrator’s interpretation of the Plan on remand, and (2) whether the Court of Appeals properly granted deference to the District Court on the merits. Pet. for Cert. i. We granted .certiorari on both, 557 U. S. 933 (2009), but find it necessary to decide only the first.
*512II
A
This Court addressed the standard for reviewing the decisions of ERISA plan administrators in Firestone, 489 U. S. 101. Because ERISA’s text does not directly resolve the matter, we looked to “principles of trust law” for guidance. Id., at 109, 111. We recognized that, under trust law, the proper standard of review of a trustee's decision depends on the language of the instrument creating the trust. See id., at 111-112. If the trust documents give the trustee “power to construe disputed or doubtful terms,... the trustee’s interpretation will not be disturbed if reasonable.” Id., at 111. Based on these considerations, we held that “a denial of benefits challenged under § 1132(a)(1)(B) is to be reviewed under a de novo standard unless the benefit plan gives the administrator or fiduciary discretionary authority to determine eligibility for benefits or to construe the terms of the plan.” Id., at 115.
We expanded Firestone’s approach in Metropolitan Life Ins. Co. v. Glenn, 554 U. S. 105 (2008). In determining the proper standard of review when a plan administrator operates under a conflict of interest, we again looked to trust law, the terms of the plan at issue, and the principles of ERISA— plus, of course, our precedent in Firestone. See 554 U. S., at 110-116. We held that, when the terms of a plan grant discretionary authority to the plan administrator, a deferential standard of review remains appropriate even in the face of a conflict. See id., at 115-116.
It is undisputed that, under Firestone and the terms of the Plan, the Plan Administrator here would normally be entitled to deference when interpreting the Plan. See 328 F. Supp. 2d, at 430-431 (observing that the Plan grants the Plan Administrator “broad discretion in making decisions relative to the Plan”). The Court of Appeals, however, crafted an exception to Firestone deference. Specifically, *513the Second Circuit held that a court need not apply a deferential standard “where the administrator ha[s] previously construed the same [plan] terms and we found such a construction to have violated ERISA.” 535 F. 3d, at 119. Under that view, the District Court here was entitled to reject a reasonable interpretation of the Plan offered by the Plan Administrator, solely because the Court of Appeals had overturned a previous interpretation by the Administrator. Cf. ibid, (accepting the District Court’s chosen method as one of “several reasonable alternatives”).
B
We reject this “one-strike-and-you’re-out” approach. Brief for Petitioners 51. As an initial matter, it has no basis in the Court’s holding in Firestone, which set out a broad standard of deference without any suggestion that the standard was susceptible to ad hoc exceptions like the one adopted by the Court of Appeals. See 489 U. S., at 111, 115. Indeed, we refused to create such an exception to Firestone deference in Glenn, recognizing that ERISA law was already complicated enough without adding “special procedural or evidentiary rules” to the mix. 554 U. S., at 116. If, as we held in Glenn, a systemic conflict of interest does not strip a plan administrator of deference, see id., at 115, it is difficult to see why a single honest mistake would require a different result.
Nor is the Court of Appeals’ decision supported by the considerations on which our holdings in Firestone and Glenn were based — namely, the terms of the plan, principles of trust law, and the purposes of ERISA. See supra, at 512. First, the Plan here grants the Plan Administrator general authority to “[c]onstrue the Plan.” App. to Pet. for Cert. 141a-142a. Nothing in that provision suggests that the grant of authority is limited to first efforts to construe the Plan.
*514Second, the Court of Appeals’ exception to Firestone deference is not required by principles of trust law. Trust law is unclear on the narrow question before us. A leading treatise states that a court will strip a trustee of his discretion when there is reason to believe that he will not exercise that discretion fairly — for example, upon a showing that the trustee has already acted in bad faith:
“If the trustee’s failure to pay a reasonable amount [to the beneficiary of the trust] is due to a failure to exercise [the trustee’s] discretion honestly and fairly, the court may well fix the amount [to be paid] itself. On the other hand, if the trustee’s failure to provide reasonably for the beneficiary is due to a mistake as to the trustee’s duties or powers, and there is no reason to believe the trustee will not fairly exercise the discretion once the court has determined the extent of the trustee’s duties and powers, the court ordinarily will not fix the amount but will instead direct the trustee to make reasonable provision for the beneficiary’s support.” 3 A. Scott, W. Fratcher, & M. Ascher, Scott and Ascher on Trusts § 18.2.1, pp. 1348-1349 (5th ed. 2007) (hereinafter Scott and Ascher) (citing cases; footnote omitted).
This is not surprising — if the settlor who creates a trust grants discretion to the trustee, it seems doubtful that the settlor would want the trustee divested entirely of that discretion simply because of one good-faith mistake.
*515Here the lower courts made no finding that the Plan Administrator had acted in bad faith or would not fairly exercise his discretion to interpret the terms of the Plan. Thus, if the District Court had followed the trust law principles set out in Scott and Ascher, it should not have “act[ed] as a substitute trustee,” Eaton v. Eaton, 82 N. H. 216, 218, 132 A. 10, 11 (1926), and stripped the Plan Administrator of the deference he would otherwise enjoy under Firestone and the terms of the Plan.
Other trust law sources, however, point the other way. For example, the Restatement (Second) of Trusts states that “the court will control the trustee in the exercise of a power where he acts beyond the bounds of a reasonable judgment.” 1 Restatement (Second) of Trusts § 187, Comment i, p. 406 (1957). Another treatise states that, after a trustee has abused his discretion, “[s]ometimes the court decides for the trustee how he should act, either by stating the exact result it desires to achieve, or by fixing some limits on the trustee’s *516action and giving him leeway within those limits.” G. Bogert & G. Bogert, Law of Trusts and Trustees §560, p. 223 (2d rev. ed. 1980).
The unclear state of trust law on the question was perhaps best captured by the Texas Supreme Court:
“There is authority for ordering a dismissal of the case to afford the trustee an opportunity to exercise a reasonable discretion in arriving at the amount of payments to be made in the light of our discussion of the problem and after a proper consideration of the many factors involved. On the other hand, there is authority for remanding the case to the trial court to hear evidence and in the exercise of its supervisory jurisdiction to fix the amount of such payments. There is still other authority for remanding the case to the trial court to hear evidence and fix the boundaries of a reasonable discretion to be exercised by the trustee within maximum and minimum limits.” State v. Rubion, 158 Tex. 43, 54-55, 308 S. W. 2d 4, 11 (1957) (citations omitted).
While we are “guided by principles of trust law” in ERISA cases, Firestone, 489 U. S., at 111, we have recognized before that “trust law does not tell the entire story,” Varity Corp. v. Howe, 516 U. S. 489, 497 (1996); see ibid. (“In some instances, trust law will offer only a starting point, after which courts must go on to ask whether, or to what extent, the language of the statute, its structure, or its purposes require departing from common-law trust requirements”); Brief for Respondents 50 (pressing same view as the dissent but concluding that the dispute over trust law “need not be resolved”). Here trust law does not resolve the specific issue before us, but the guiding principles we have identified underlying ERISA do.
Congress enacted ERISA to ensure that employees would receive the benefits they had earned, but Congress did not require employers to establish benefit plans in the first place. *517Lockheed Corp. v. Spink, 517 U. S. 882, 887 (1996). We have therefore recognized that ERISA represents a “ ‘careful balancing’ between ensuring fair and prompt enforcement of rights under a plan and the encouragement of the creation of such plans.” Aetna Health Inc. v. Davila, 542 U. S. 200, 215 (2004) (quoting Pilot Life Ins. Co. v. Dedeaux, 481 U. S. 41, 54 (1987)). Congress sought “to create a system that is [not] so complex that administrative costs, or litigation expenses, unduly discourage employers from offering [ERISA] plans in the first place.” Varity Corp., supra, at 497. ERISA “induc[es] employers to offer benefits by assuring a predictable set of liabilities, under uniform standards of primary conduct and a uniform regime of ultimate remedial orders and awards when a violation has occurred.” Rush Prudential HMO, Inc. v. Moran, 536 U. S. 355, 379 (2002).
Firestone deference protects these interests and, by permitting an employer to grant primary interpretive authority over an ERISA plan to the plan administrator, preserves the “careful balancing” on which ERISA is based. Deference promotes efficiency by encouraging resolution of benefits disputes through internal administrative proceedings rather than costly litigation. It also promotes predictability, as an employer can rely on the expertise of the plan administrator rather than worry about unexpected and inaccurate plan interpretations that might result from de novo judicial review. Moreover, Firestone deference serves the interest of uniformity, helping to avoid a patchwork of different interpretations of a plan, like the one here, that covers employees in different jurisdictions — a result that “would introduce considerable inefficiencies in benefit program operation, which might lead those employers with existing plans to reduce benefits, and those without such plans to refrain from adopting them.” Fort Halifax Packing Co. v. Coyne, 482 U. S. 1, 11 (1987). Indeed, a group of prominent actuaries tells us that it is impossible even to determine whether an ERISA plan is solvent (a duty imposed on actuaries by federal law, *518see 29 U. S. C. §§ 1023(a)(4), (d)) if the plan is interpreted to mean different things in different places. See Brief for Chief Actuaries as Amici Curiae 5-11.
Respondents and the United States as amicus curiae do not question that deference to plan administrators serves these important purposes. Rather, they argue that deference is less important once a plan administrator has issued an interpretation of a plan found to be unreasonable. But the interests in efficiency, predictability, and uniformity— and the manner in which they are promoted by deference to reasonable plan construction by administrators — do not suddenly disappear simply because a plan administrator has made a single honest mistake.
This case illustrates the point. Consider first the interest in efficiency, an interest that Xerox has pursued by granting the Plan Administrator authority to construe the Plan. On remand from the Court of Appeals, if the District Court had applied a deferential standard of review under Firestone, the question before it would have been whether the Plan Administrator’s interpretation of the Plan was reasonable. After answering that question, the case might well have been over. Instead, the District Court declined to defer, and therefore had to answer the more complicated question of how best to interpret the Plan.
The prospect of increased litigation costs inherent in respondents’ approach does not end there. Under respondents’ and the Government’s view, the question whether a deferential standard of review was required in this case turns on whether the Plan Administrator was interpreting the “same terms” or deciding the “same issue” on remand. See Brief for Respondents 43, 46-48, 53, and n. 13; Brief for United States as Amicus Curiae 13-15, 23. Whether that condition is satisfied will not always be clear. Indeed, petitioners dispute that question here, arguing that the Plan Administrator confronted an entirely new issue on remand— how to interpret the Plan, knowing that specific provisions *519requiring use of the phantom account method could not be applied to respondents due to a lack of notice. See Brief for Petitioners 50-51. Respondents would force the parties to litigate this potentially complicated “same issue” or “same terms” question before a district court could even decide whether deference is owed to a plan administrator’s view. As we recognized in Glenn, there is little place in the ERISA context for these sorts of “special procedural rules [that] would create further complexity, adding time and expense to a process that may already be too costly for many of those who seek redress.” 554 U. S., at 116-117.
The position of respondents and the Government could interject other additional issues into ERISA litigation. For example, even under their view, the District Court here could have granted deference to the Plan Administrator; the court merely was not required to do so. See Brief for Respondents 43, 49-50, 52-53; Brief for United States as Amicus Curiae 23-24. That raises the question of how a court is to decide between the two options; respondents’ answer is to weigh an indeterminate number of factors, which would only further complicate ERISA proceedings. See Tr. of Oral Arg. 34, 40-45.
This ease also demonstrates the harm to the interest in predictability that would result from stripping a plan administrator of Firestone deference. After declining to apply a deferential standard here, the District Court adopted an interpretation of the Plan that does not account for the time value of money. 472 F. Supp. 2d, at 458; 535 F. 3d, at 119. In the actuarial world, this is heresy, and highly unforeseeable. Indeed, the actuaries tell us that they have never encountered an ERISA plan resembling this one that did not include some adjustment for the time value of money. Brief for Chief Actuaries as Amici Curiae 12.
Respondents’ own actuarial expert testified before the District Court that fairness would require recognizing the time value of money in some fashion. See App. 127a, 130a. *520And respondents and the Government do not dispute that the District Court’s approach, which does not account for the fact that respondents were able to use their past distributions as they saw fit for over 20 years, would place respondents in a better position than employees who never left the company. Cf. Brief for Respondents 42-43; Brief for United States as Amicus Curiae 32-33. Deference to plan administrators, who have a duty to all beneficiaries to preserve limited plan assets, see Varity Corp., 516 U. S., at 514, helps prevent such windfalls for particular employees.
Finally, this case demonstrates the uniformity problems that arise from creating ad hoc exceptions to Firestone deference. If other courts were to adopt an interpretation of the Plan that does account for the time value of money, Xerox could be placed in an impossible situation. Similar Xerox employees could be entitled to different benefits depending on where they live, or perhaps where they bring a legal action. Cf. 29 U. S. C. § 1132(e)(2) (permitting suit “where the plan is administered, where the breach took place, or where a defendant resides or may be found”). In fact, that may already be the case. In similar litigation over the Plan, the Ninth Circuit also rejected the use of the phantom account method, but held that the Plan Administrator should utilize actuarial principles in accounting for rehired employees’ past distributions — which would presumably include taking some cognizance of the time value of money. See Miller v. Xerox Corp. Retirement Income Guarantee Plan, 464 F. 3d 871, 875-876 (2006); Brief for ERISA Industry Committee et al. as Amici Curiae 8-9. Thus, failing to defer to the Plan Administrator here could well cause the Plan to be subject to different interpretations in California and New York. “Uniformity is impossible, however, if plans are subject to different legal obligations in different States.” Egelhoff v. Egelhoff, 532 U. S. 141, 148 (2001). Firestone deference serves to avoid that result and to preserve the *521“careful balancing” of interests that ERISA represents. Pilot Life Ins. Co., 481 U. S., at 54.
C
In spite of all this, respondents and the Government argue that requiring the District Court to apply Firestone deference in this case would actually disserve the purposes of ERISA. They argue that continued deference would encourage plan administrators to adopt unreasonable interpretations of plans in the first instance, as administrators would anticipate a second chance to interpret their plans if their first interpretations were rejected. And they argue that plan administrators would be able to proceed seriatim through several interpretations of their plans, each time receiving deference, thereby undermining the prompt resolution of disputes over benefits, driving up litigation costs, and discouraging employees from challenging the decisions of plan administrators at all.
All this is overblown. There is no reason to think that deference would be required in the extreme circumstances that respondents foresee. Under trust law, a trustee may be stripped of deference when he does not exercise his discretion “honestly and fairly.” 3 Scott and Ascher 1348. Multiple erroneous interpretations of the same plan provision, even if issued in good faith, might well support a finding that a plan administrator is too incompetent to exercise his discretion fairly, cutting short the rounds of costly litigation that respondents fear.
Applying a deferential standard of review does not mean that the plan administrator will prevail on the merits. It means only that the plan administrator's interpretation of the plan “will not be disturbed if reasonable.” Firestone, 489 U. S., at 111; see also ibid. (“ Where discretion is conferred upon the trustee with respect to the exercise of a power, its exercise is not subject to control by the court except to prevent an abuse by the trustee of his discretion’ ” *522(quoting 1 Restatement (Second) of Trusts § 187)). Thus, far from “impos[ing] [a] rigid and inflexible requirement” that courts must defer to plan administrators, post, at 529, we simply hold that the lower courts should have applied the standard established in Firestone and Glenn.
III
The Court of Appeals erred in holding that the District Court could refuse to defer to the Plan Administrator’s interpretation of the Plan on remand, simply because the Court of Appeals had found a previous related interpretation by the Administrator to be invalid. Because we reverse on that ground, we do not reach the question whether the Court of Appeals also erred in applying a deferential standard of review to the decision of the District Court on the merits.
The judgment of the Court of Appeals for the Second Circuit is reversed, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.
Justice Sotomayor took no part in the consideration or decision of this ease.
Justice Breyer,
with whom Justice Stevens and Justice Ginsburg join, dissenting.
I agree with the Court that “[pjeople make mistakes,” ante, at 509, but I do not share its view of the law applicable to those mistakes. To explain my view, I shall describe the three significant mistakes involved in this case.
*523I
A
The first mistake is that of Xerox Corporation’s pension plan (Plan) and its administrators (collectively, Plan Administrator or Administrator), petitioners here. The Plan, as I understand it, pays employees the highest of three benefits upon retirement. App. 29a-31a. These benefits are calculated as follows (I simplify and use my own words, not those of the Plan):
(1) “The Pension”: Take your average salary for your five highest salary years at Xerox; multiply by 1.4 percent; and multiply again by the number of years you worked at Xerox (up to 30). Id., at 7a-lla, 29a-30a. Thus, if the average salary of your five highest paid years was $50,000 and you worked at Xerox for 30 years, you would be entitled to receive $21,000 per year ($50,000 x 1.4 percent x 30).
(2) “The Cash Account”: Every year, Xerox credits 5 percent of your salary to a cash account. Id., at 40a. This account accrues interest at a yearly fixed rate 1 percent above the 1-year Treasury bill rate. Id., at 41a. To determine your benefits under this approach, take the balance of your cash account, and convert the final amount to an annuity. Id., at 31a. Thus, if you have accrued, say, $200,000 in your account, and the relevant annuity rate at the time of your retirement is 7 percent, you would be entitled to receive approximately $14,000 per year upon your retirement (approximately $200,000 x 7 percent).
(3) “The Investment Account”: Before 1990, Xerox contributed to an employee profit-sharing plan. Id., at 33a-34a. Thus, all employees who were hired by the end of 1989 have an investment account that consists of all of the contributions Xerox made to this profit-sharing plan (prior to its discontinuation) and the investment *524returns on those contributions. Id., at 33a-36a. To determine your benefits under this approach, take the balance of your investment account, and convert the final amount to an annuity. Id., at 31a. Thus, just like the cash account, if you have accrued $400,000 in your account, and the relevant annuity rate at the time of your retirement is 7 percent, you would be entitled to receive approximately $28,000 per year upon your retirement (approximately $400,000 x 7 percent).
Given these three examples, the retiring employee’s pension would come from the investment account, and the employee would receive $28,000 per year.
This case concerns one aspect of Xerox’s retirement plan, namely, the way in which the Plan treats employees who leave Xerox and later return, working for additional years before their ultimate retirement. The Plan has long treated such leaving-and-returning employees as follows (again, I simplify and use my own words):
First, when an employee initially leaves, she is paid a lump-sum distribution equivalent to the benefits she has accrued up to that point (i. e., the highest of her pension, her cash account, or, if she was hired before the end of 1989, her investment account). See ante, at 510.
Second, when the employee returns, she again begins to accrue amounts in her cash account, App. 40a-41a, starting from scratch. (She accrues nothing in her investment account, because Xerox no longer makes profit-sharing contributions. Id., at 34a.) Thus, by the time of her retirement the employee may not have accrued much money in this account.
Third, a rehired employee’s pension is calculated in the way I have set forth above, with her entire tenure at Xerox (both before her departure and after her return) taken into account. See Brief for Petitioners 9-10.
Fourth, the employee’s benefits calculation is adjusted to take account of the fact that the employee has already re*525ceived a lump-sum distribution from the Plan. See App. 32a; Brief for Petitioners 10-11.
This case is about the adjustment that takes place during step four. It concerns the way in which the Plan Administrator calculates that adjustment so as to reflect the fact that a retiring leaving-and-returning employee has already received a distribution when she initially left Xerox. Before 1989, the Plan Administrator calculated the adjusted amount by taking the benefits distribution previously received (say, $100,000) and adjusting it to equal the amount that would have existed in the investment account had no distribution been made. Ibid. Thus, if an employee had not left Xerox, and if the $100,000 had been left in her investment account for, say, 20 years, that amount would likely have increased dramatically — perhaps doubling, tripling, or quadrupling in amount, depending upon how well the Plan’s investments performed.
It is this hypothetical sum — termed the “phantom account,” ante, at 510 — that is at issue in this case. Xerox’s pre-1989 Plan assumed that a rehired employee had this hypothetical sum on hand at the time of her final retirement from the company, and in effect subtracted the amount from the employee’s benefits upon her departure. Brief for Petitioners 10-11; cf. ante, at 510. Depending on how the Plan’s investments did over time, the Administrator’s use of this “phantom account” could have a substantial impact on a rehired employee’s benefits. (See Appendix, infra, for an example of how this “phantom account” works.)
When the Plan Administrator amended Xerox’s Employee Retirement Income Security Act of 1974 (ERISA) Plan in 1989, however, it made what it tells us was an “inadverten[t]” omission. Brief for Petitioners 11, n. 3. In a section of the 1989 Plan applicable to the roughly 100 leaving-and-returning employees who are plaintiffs here, the Plan said that it would “offset” the retiring employees’ “accrued benefit” (as ordinarily calculated) “by the accrued benefit attrib*526utable” to the prior lump-sum “distribution” those employees received when they initially left Xerox. App. 32a. But the Plan said nothing about how it would calculate this “offset. ” In other words, the Plan said nothing about the Administrator’s use of the “phantom account.”
This led to the first mistake in this case. Despite the Plan’s failure to include language explaining how the Administrator would take into account an employee’s prior distribution, the Plan Administrator continued to employ the “phantom account” methodology. In essence, the Administrator read the 1989 Plan to include the language that had been omitted — an interpretation that, as described below, see Part I-B, infra, the Court of Appeals found to be arbitrary and capricious and in violation of ERISA.
B
The District Court committed the second mistake in this case. In 1999, respondents, nearly 100 employees who left and were later rehired by Xerox, brought this lawsuit. Ante, at 510; Brief for Petitioners ii-iii, 12. They pointed out that the 1989 Plan said that it would decrease their retirement benefits to reflect the fact that they had already received a lump-sum benefits distribution when they initially left Xerox. But, they added, neither the 1989 Plan, nor the 1989 Plan’s Summary Plan Description, said anything about whether (or how) the Administrator would adjust their previous benefits distribution to take into account that they had received the distribution well before their retirement. They thus claimed that the Plan Administrator could not use the “phantom account” methodology to adjust their previous distributions. See Brief for United States as Amicus Curia/e 4-5.
The District Court, however, rejected respondents’ claims. 328 F. Supp. 2d 420 (WDNY 2004). The court accepted the Administrator’s argument that the 1989 Plan implicitly incorporated the “phantom account” approach that had previously *527been part of Xerox’s retirement plan. Id., at 433-434. And the court thus held in favor of petitioners — thereby committing the second mistake in this case. Id., at 439.
On appeal, the Second Circuit disagreed with the District Court and vacated the District Court’s decision in relevant part. 433 F. 3d 254 (2006). The Court of Appeals concluded that, because the 1989 Plan said nothing about how the Administrator would adjust the previous benefits distributions, it was “arbitrary and capricious” for the Administrator to interpret the 1989 Plan as if it still incorporated the “phantom account.” Id., at 265-266, and n. 11. And the Court of Appeals thus held that the language of the Plan and the Summary Plan Description, at the least, violated ERISA by failing to provide respondents with fair notice that the Administrator was going to use the “phantom account” approach. See id., at 265 (discussing 29 U. S. C. § 1022); see also 433 F. 3d, at 263, 267-268 (holding that the Administrator’s attempt to apply the “phantom account” to respondents violated two other ERISA provisions: 29 U. S. C. § 1054(h)’s notice requirement and § 1054(g)’s prohibition on retroactive benefit cutbacks). Rather, the court noted, respondents “likely believed” — based on the language of the Plan — “that their past distributions would only be factored into their [current] benefits calculations by taking into account the amounts they had actually received.” 433 F. 3d, at 267.
In light of these conclusions, the Court of Appeals recognized the need to devise a remedy for the Administrator’s abuse of discretion and ERISA violations — a remedy that took into account the previous benefits distributions respondents had received in a manner consistent with the 1989 Plan. The court therefore remanded the case to the District Court, with the following instructions:
“On remand, the remedy crafted by the district court for those employees [in the respondents’ situation] should utilize an appropriate [pre-1989 Plan] calculation to determine their benefits. We recognize the difficulty *528that this task poses .... As guidance for the district court, we suggest that it may wish to employ equitable principles when determining the appropriate calculation and fashioning the appropriate remedy.” Id., at 268.
On remand, the District Court invited the parties to submit remedial recommendations. Brief for Petitioners 14. The Plan Administrator proposed an approach that would adjust respondents’ previous benefits distributions by adding interest, and, as a fallback, the Administrator suggested that the Plan should treat respondents as new hires. Ante, at 510-511; Brief for United States as Amicus Curiae 6-7. The District Court rejected these suggestions and concluded that the “appropriate” remedy was the one suggested by the Second Circuit: no adjustment to the prior distributions received by respondents. 472 F. Supp. 2d 452, 458 (WDNY 2007). The court stated that this remedy was “straightforward; it adequately prevent[ed] employees from receiving a windfall[;] and... it most clearly refleet[ed] what a reasonable employee would have anticipated based on the not-very-clear language in the Plan.” Ibid. And the Court of Appeals, finding that the District Court did not abuse its discretion in crafting a remedy, affirmed. 535 F. 3d 111 (CA2 2008).
II
The third mistake, I believe, is the Court’s. As the majority recognizes, ante, at 512, “principles of trust law” guide this Court in “determining the appropriate standard” by which to review the actions of an ERISA plan administrator. Firestone Tire & Rubber Co. v. Bruch, 489 U. S. 101, 111-113 (1989); see also Metropolitan Life Ins. Co. v. Glenn, 554 U. S. 105, 111 (2008); Aetna Health Inc. v. Davila, 542 U. S. 200, 218-219 (2004); Central States, Southeast & Southwest Areas Pension Fund v. Central Transport, Inc., 472 U. S. 559, 570 (1985). And, as the majority also recognizes, ante, at 512, where an ERISA plan grants an administrator the discretionary authority to interpret plan terms, trust law requires *529a court to defer to the plan administrator’s interpretation of plan terms. See, e. g., Glenn, supra, at 111. But the majority further concludes that trust law “does not resolve the specific issue before” the Court in this case — i. e., whether a court is required to defer to an administrator’s second attempt at interpreting plan documents, even after the court has already determined that the administrator’s first attempt amounted to an abuse of discretion. Ante, at 516. In my view, this final conclusion is erroneous, as trust law imposes no such rigid and inflexible requirement.
The Second Circuit found the Administrator’s interpretation of the Plan to be arbitrary and capricious and in violation of ERISA, and it made clear that the District Court's task on remand was to “craf [t]” a “remedy.” See 433 F. 3d, at 268. Trust law treatise writers say that in these circumstances a court may (but need not) exercise its own discretion rather than defer to a trustee’s interpretation of trust language. See G. Bogert & G. Bogert, Law of Trusts and Trustees §560, pp. 222-223 (2d rev. ed. 1980) (hereinafter Bogert & Bogert) (after finding an abuse of discretion, a court may “decid[e] for the trustee how he should act,” possibly by “stating the exact result” the court “desires to achieve”); see also 2 Restatement (Third) of Trusts §50, p. 258 (2001) (hereinafter Third Restatement) (“A discretionary power conferred upon the trustee ... is subject to judicial control only to prevent misinterpretation or abuse of the discretion by the trustee”); 1 Restatement (Second) of Trusts § 187, p. 402 (1957) (hereinafter Second Restatement) (“Where discretion is conferred upon the trustee ... , its exercise is not subject to control by the court, except to prevent an abuse by the trustee of his discretion”); see also Firestone, supra, at 111. Judges deciding trust law cases have said the same. See, e. g., Colton v. Colton, 127 U. S. 300, 322 (1888) (stating that it was the “duty of the court” to determine the trust payments due after rejecting the trustee’s interpretation); State v. Rubion, 158 Tex. 43, 55, 308 S. W. *5302d 4, 11 (1957) (“Considering that we have held that there has already been an abuse of discretion by the trustee ..., we have concluded that a remand of the case to the trial court for the definite establishment of amounts to be paid will better promote a speedy administration of justice and a final termination of this litigation”); Glenn, supra, at 130 (Scalia, J., dissenting) (court may exercise discretion under trust law when a “trustee had discretion but abused it”). In short, the controlling trust law principle appears to be that, “[w]here the court finds that there has been an abuse of a discretionary power, the decree to be rendered is in its discretion.” Bogert & Bogert § 560, at 222.
Of course, the fact that trust law grants courts discretion does not mean that they will exercise that discretion in all instances. The majority refers to the 2007 edition of Scott on Trusts, ante, at 514, which says that, if there is “no reason” to doubt that a trustee “will . . . fairly exercise” his “discretion,” then courts “ordinarily will not fix the amount” of a payment “but will instead direct the trustee to make reasonable provision for the beneficiary’s support,” 3 A. Scott, W. Fratcher, & M. Ascher, Scott and Ascher on Trusts § 18.2.1, pp. 1348-1349 (5th ed. 2007) (hereinafter Scott) (emphasis added). As this passage demonstrates, there are situations in which a court will typically defer to a trustee’s remedial suggestion. The word “ordinarily” confirms, however, that the Scott treatise writers recognize that there are instances in which courts will not defer. And other treatises indicate that black letter trust law gives the district courts authority to decide which instances are which. See Bogert & Bogert § 560, at 222-223 (when there is an abuse of discretion, a court “may set aside the transaction,” “award damages to the beneficiary,” or “order a new decision to be made in the light of rules expounded by the court”); 2 Third Restatement § 50, and Comment b, at 261 (discussing similar remedial options); 1 Second Restatement §187, and Comment b, at 402 (same); see also 3 Third Restatement §87, and *531Comment c, at 244-245 (noting that “judicial intervention on the ground of abuse” is allowed when a “good faith,” yet “unreasonable,” decision is made by a trustee); Rubion, supra, at 54-55, 308 S. W. 2d, at 11 (discussing a court’s remedial options).
Nevertheless, the majority reads the Scott treatise as establishing an absolute requirement that courts defer to a trustee’s fallback position absent “reason to believe that [the trustee] will not exercise [his] discretion fairly — for example, upon a showing that the trustee has already acted in bad faith.” Ante, at 514. And based on this reading, the majority further concludes that the existence of the Scott treatise creates uncertainty as to whether, under basic trust law principles, a court has the power to craft a remedy for a trustee’s abuse of discretion. Ante, at 514-516.
It is unclear to me, however, why the majority reads the passage from Scott as creating a war among treatise writers, compare ante, at 514 (discussing Scott), with ante, at 515-516 (discussing Bogert), when the relevant passages can so easily be read as consistent with one another. I simply read the Scott treatise language as identifying circumstances in which courts typically choose to defer to an administrator’s fallback position. The treatise does not suggest that the law prohibits a court from acting on its own in the exercise of its broad remedial authority — authority that trust law plainly grants to supervising courts. See supra, at 530.
A closer look at the Scott treatise confirms this understanding. The treatise cites seven cases in support of the passage upon which the majority relies. See 3 Scott § 18.2.1, at 1349, n. 4. Three of these cases explicitly state that a court may exercise its discretion to craft a remedy if a trustee has previously abused its discretion. See Old Colony Trust Co. v. Rodd, 356 Mass. 584, 589, 254 N. E. 2d 886, 889 (1970) (“A court of equity may control a trustee in the exercise of a fiduciary discretion if it fails to observe standards of judgment apparent from the applicable instrument”); *532In re Estate of Marré, 18 Cal. 2d 184, 190, 114 P. 2d 586, 590-591 (1941) (“It is well settled that the courts will not attempt to exercise discretion which has been confided to a trustee unless it is clear that the trustee has abused his discretion in some manner” (emphasis added)); In re Estate of Ferrall, 92 Cal. App. 2d 712, 716-717, 207 P. 2d 1077, 1079-1080 (1949) (following In re Estate of Marré). Three other cases are inapposite because their circumstances do not involve any allegation of abuse of discretion by the trustee. See In re Trusts of Ziegler, 157 So. 2d 549, 550 (Fla. App. 1963) (per curiam) (“There is no contention here that the court... would not retain its rights, upon appropriate petition or other pleadings by an interested party, to review an alleged abuse, if any, of the discretion exercised by the trustees”); In re Estate of Grubel, 37 Misc. 2d 910, 911, 235 N. Y. S. 2d 21, 23 (Surr. Ct. 1962) (stating that “in the first instance” it is the “proper function of the trustees” to set an amount to be paid (emphasis added)); Orr v. Moses, 94 N. H. 309, 312, 52 A. 2d 128, 130 (1947) (declining to construe will because none “of the parties now assert claims adverse to any position taken by the trustee”). In the final case, the court decided that, on the facts before it, it did not need to control the trustees’ discretion. See In re Estate of Stillman, 107 Misc. 2d 102, 111, 433 N. Y. S. 2d 701, 708 (Surr. Ct. 1980) (“The fine record of the trustees in enhancing the equity of these trusts while earning substantial income, also persuades the court of the wisdom of retaining their services as fiduciaries”). Which of these cases says that, after the trustee has abused its discretion, a district court must still defer to the trustee? None of them do. I repeat: Not a single case cited by the Scott treatise writers supports the majority’s reading of the treatise.
The majority seeks to justify its reading of the Scott treatise by referring to four cases that Scott does not cite. See ante, at 514-515, n. 1. I am not surprised that the treatise does not refer to these cases. In the first three, a court *533thought it best, when a trustee had not yet exercised judgment about a particular matter, to direct the trustee to do so. See In re Will of Sullivan, 144 Neb. 36, 40-41,12 N. W. 2d 148, 150-151 (1943) (finding that the trustees’ “failure to act” was erroneous, and directing the trustees to exercise their discretion in setting a payment amount); Eaton v. Eaton, 82 N. H. 216, 218, 132 A. 10, 11 (1926) (same); Finch v. Wachovia Bank & Trust Co., 156 N. C. App. 343, 347-348, 577 S. E. 2d 306, 309-310 (2003) (holding trustee erred by “[f]ail[ing] to exercise judgment,” and directing it to do so). The fourth case concerns circumstances so distant from those before us that it is difficult to know what to say. (The question was whether the beneficiary of a small trust had title in certain trust assets or whether the trustee had discretionary power to allocate them in her best interest; the court held the latter, adding that, if the trustee acted unreasonably, the lower court in that particular case should seek to have the trustee removed rather than trying to administer the trust funds itself.) See Hanford v. Clancy, 87 N. H. 458, 460-461, 183 A. 271, 272-273 (1936).
I cannot read these four cases, or any other case to which the majority refers, as holding that a court, as a general matter, is required to defer to a trust administrator’s second attempt at exercising discretion. And I am aware of no such case. In contrast, the Restatement and Bogert and Scott treatises identify numerous cases in which courts have remedied a trustee’s abuse of discretion by ordering the trustee to pay a specific amount. See 2 Third Restatement § 50, Reporter’s Note, at 283 (citing cases such as Coker v. Coker, 208 Ala. 354, 94 So. 566 (1922)); Bogert & Bogert § 560, at 223, n. 19 (citing cases such as Rubion); 3 Scott § 18.2.1, at 1348-1349, nn. 3-4 (citing cases such as Emmert v. Old Nat. Bank of Martinsburg, 162 W. Va. 48, 246 S. E. 2d 236 (1978)); see also Brief for United States as Amicus Curiae 18 (listing cases). I thus do not find trust law “unclear” on this matter. Ante, at 514. When a trustee abuses its discretion, trust law *534grants courts the authority either to defer anew to the trustee’s discretion or to craft a remedy. See, e.g., 3 A. Scott & W. Fratcher, Scott on Trusts § 187, pp. 14-15 (4th ed. 1988) (“This ordinarily means that so long as [the trustee] acts not only in good faith and from proper motives, but also within the bounds of reasonable judgment, the court will not interfere; but the court will interfere when he acts outside the bounds of a reasonable judgment”).
Nor does anything in the present case suggest that the District Court abused its remedial authority. The Second Circuit stated that the interpretive problem on remand was in essence a remedial problem. See 433 F. 3d, at 268. It added that the remedial problem was “difficult]’’ and that “the district court... may wish to employ equitable principles when determining the appropriate calculation and fashioning the appropriate remedy.” Ibid. The Administrator had previously abused his discretionary power. Id., at 265-268. And the District Court found that the Administrator’s primary remedial suggestion on remand — adjusting respondents’ previous benefits distributions by adding interest— probably would have violated ERISA’s notice provisions. 472 F. Supp. 2d, at 457. Under these circumstances, the District Court reasonably could have found a need to use its own remedial judgment, rather than rely on the Administrator’s — which is just what the Second Circuit said. 535 F. 3d, at 119.
Moreover, even if the “narrow” trust law “question before us” were difficult, ante, at 514 — which it is not — this difficulty would not excuse the Court from trying to do its best to work out a legal solution that nonetheless respects basic principles of trust law. “Congress invoked the common law of trusts” in enacting ERISA, and this Court has thus repeatedly looked to trust law in order to determine “the particular duties and powers” of ERISA plan administrators. Central States, Southeast & Southwest Areas Pension Fund v. Central Transport, Inc., 472 U. S., at 570-572; see also, *535e. g., Glenn, 554 U. S., at 111; Davila, 542 U. S., at 218-219; Firestone, 489 U. S., at 111-113. While, as the majority recognizes, ante, at 516, trust law may “not tell the entire story,” Varity Corp. v. Howe, 516 U. S. 489, 497 (1996), I am aware of no other case in which this Court has simply ignored trust law (on the basis that it was unclear) and crafted a legal rule based on nothing but “the guiding principles we have identified underlying ERISA,” ante, at 516. See Varity, supra, at 497 (“In some instances, trust law will offer only a starting point, after which courts must go on to ask whether, or to what extent, the language of the statute, its structure, or its purposes require departing from common-law trust requirements” (emphasis added)).
In any event, it is far from clear that the Court’s legal rule reflects an appropriate analysis of ERISA-based policy. To the contrary, the majority’s absolute “one free honest mistake” rule is impractical, for it requires courts to determine what is “honest,” encourages appeals on the point, and threatens to delay further proceedings that already take too long. (Respondents initially filed this retirement benefits case in 1999.) See Glenn, 554 U. S., at 116-117. It also ignores what we previously have pointed out — namely, that abuses of discretion “arise in too many contexts” and “concern too many circumstances” for this Court “to come up with a one-size-fits-all procedural [approach] that is likely to promote fair and accurate” benefits determinations. Ibid. And, finally, the majority’s approach creates incentives for administrators to take “one free shot” at employer-favorable plan interpretations and to draft ambiguous retirement plans in the first instance with the expectation that they will have repeated opportunities to interpret (and possibly reinterpret) the ambiguous terms. I thus fail to see how the majority’s “one free honest mistake” approach furthers ERISA’s core purpose of “promoting] the interests of employees and their beneficiaries in employee benefit plans.” Shaw v. Delta Air Lines, Inc., 463 U. S. 85, 90 (1983); see also, e. g., *53629 U. S. C. § 1001(b) (noting that ERISA was enacted “to protect . . . employee benefit plans and their beneficiaries”); Curtiss-Wright Corp. v. Schoonejongen, 514 U. S. 73, 83 (1995) (discussing ERISA’s central “goa[l]” of “enabling] plan beneficiaries to learn their rights and obligations at any time”); Massachusetts Mut. Life Ins. Co. v. Russell, 473 U. S. 134, 148 (1985) (ERISA was enacted “to protect contractually defined benefits”).
The majority does identify ERISA-related factors — e.g., promoting predictability and uniformity, encouraging employers to adopt strong plans — that it believes favor giving more power to plan administrators. See ante, at 517-521. But, in my view, these factors are, at the least, offset by the factors discussed above — e. g., discouraging administrators from writing opaque plans and interpreting them aggressively — that argue to the contrary. At best, the policies at issue — some arguing in one direction, some the other — are far less able than trust law to provide a “guiding principle.” Thus, I conclude that here, as elsewhere, trust law ultimately provides the best way for courts to approach the administration and interpretation of ERISA. See, e.g., Firestone, supra, at 111-113. And trust law here, as I have said, leaves to the supervising court the decision as to how much weight to give to a plan administrator’s remedial opinion.
III
Since the District Court was not required to defer to the Administrator’s fallback position, I should consider the second question presented, namely, whether the Court of Appeals properly reviewed the District Court’s decision under an “abuse of discretion” standard. Ante, at 511 (acknowledging, but not reaching, this issue). The answer to this question depends upon how one characterizes the Court of Appeals’ decision. If the court deferred to the District Court’s interpretation of Plan terms, then the Court of Appeals most likely should have reviewed the decision de novo. See Fire*537stone, supra, at 112; cf. Davila, supra, at 210 (“Any dispute over the precise terms of the plan is resolved by a court under a de novo review standard”). If instead the Court of Appeals deferred to the District Court’s creation of a remedy, in significant part on the basis of “equitable principles,” then it properly reviewed the District Court decision for “abuse of discretion.” See, e. g., Cook v. Liberty Life Assurance Co. of Boston, 320 F. 3d 11, 24 (CA1 2003); Zervos v. Verizon N. Y., Inc., 277 F. 3d 635, 648 (CA2 2002); Grosz-Salomon v. Paul Revere Life Ins. Co., 237 F. 3d 1154, 1163 (CA9 2001); Halpin v. W. W. Grainger, Inc., 962 F. 2d 685, 697 (CA7 1992).
The District Court opinion contains language that supports either characterization. On the one hand, the court wrote that its task was to “interpret the Plan as written.” 472 F. Supp. 2d, at 457. On the other hand, the court said that “virtually nothing is set forth in either the Plan or the [Summary Plan Description]” about how to treat prior distributions; and, in describing its task, it said that the Court of Appeals had directed it to use “equitable principles” in fashioning a remedy. Ibid. Ultimately, the District Court appears to have used both the Plan language and equitable principles to arrive at its conclusion. See id., at 457-459.
The Court of Appeals, too, used language that supports both characterizations. Compare 535 F. 3d, at 117 (noting that the District Court “applied [Plan] terms” in crafting its remedy), with id., at 117-119 (describing the District Court’s decision as the “eraft[ing]” of a “remedy” and acknowledging that it had directed the District Court to use “equitable principles” in doing so). But the Court of Appeals ultimately treated the District Court’s opinion as if it primarily created a fair remedy. Ibid. Given the prior Court of Appeals opinion’s language, supra, at 527-528 (quoting 433 F. 3d, at 268), I believe that view is a fair, indeed a correct, view. And I consequently believe the Court of Appeals properly reviewed the result for an “abuse of discretion.”
*538Petitioners argue that, because respondents were seeking relief under 29 U. S. C. § 1132(a)(1)(B), the Court of Appeals was, in effect, prohibited from treating the remedy as anything other than an application of a plan’s terms. Brief for Petitioners 55-56; Reply Brief for Petitioners 3, 16-17, and n. 8. While this provision allows plaintiffs only to “enforce” or “clarify” rights or to “recover benefits” “under the terms of the plan,” § 1132(a)(1)(B) (emphasis added), it does not so limit a court’s remedial authority, Great-West Life & Annuity Ins. Co. v. Knudson, 534 U. S. 204, 221 (2002) (In § 1132(a)(1)(B), “Congress authorized ‘a participant or beneficiary’ to bring a civil action . . . without referencfing] whether the relief sought is legal or equitable”). The provision thus does not prohibit a court from shaping relief through the application of equitable principles, as trust law plainly permits. See, e. g., 2 Third Restatement § 50, and Comment b, at 261 (discussing remedial options); Bogert & Bogert §870, at 123-126 (2d rev. ed. 1995). Indeed, a court that finds, for example, that an administrator provided employees with inadequate notice of a plan’s terms (as was true here) may have no alternative but to rely significantly upon those principles. Cf. 29 U. S. C. § 1104(a)(1)(D) (plan fiduciary must “discharge his dut[y] ... in accordance with the documents and instruments governing the plan insofar as such documents and instruments are consistent” with ERISA).
For these reasons I would affirm the decision of the Court of Appeals. And I therefore respectfully dissent from the majority’s contrary determination.
APPENDIX
The “Phantom Account”
This Appendix provides a simplified and illustrative example of, as I understand it, how the “phantom account” works. For the purposes of this Appendix, I make the following assumptions: John worked at Xerox for 10 years from 1970 to *5391980. At the time of his departure from Xerox, he was issued a lump-sum benefits distribution of $140,000. He was then rehired in January 1989, and he worked for Xerox for five more years before retiring (until December 1993), earning $50,000 each year of his second term of employment. I also assume that (1) Xerox’s contribution to John’s investment account was $2,500 in 1989 (the last year such accounts were offered), (2) Xerox’s contributions to John’s cash and investment accounts are always made on the final day of the year, (3) the rate of return in John’s cash and investment accounts is always 5 percent, and (4) annuity rates are also always 5 percent. (For the sake of simplicity, I treat all annuities as perpetuities, meaning that I calculate the present value of the annuities thusly: Present Value = Annual Payment/Annuity Rate.)
Given the above assumptions, John’s pension upon his retirement would be $10,500 per year ($50,000 x 1.4 percent x 15 years), which has a present value of $210,000 ($10,500 -s- 5 percent). John’s cash and investment accounts at the end of his fifth year would look as follows (While Xerox’s ERISA Plan did not include cash accounts until 1990, each employee’s opening cash account balance was credited with the balance of his investment account at the end of 1989. The figures for John’s cash account in 1989 thus reflect the performance of his investment account. In addition, all numbers are rounded to the nearest hundred):
*540Now, as far as I understand it, John’s retirement benefits are calculated as follows, see 433 F. 3d, at 260:
First, the Plan Administrator would choose which of John’s three accounts would yield him the greatest benefits. In making this comparison, the Plan Administrator would assume that John had never left Xerox when calculating John's pension. The Plan Administrator would also assume, when calculating the value of John’s cash and investment accounts, that the lump-sum distribution John had received from Xerox had remained invested in his accounts. (In other words, the Plan Administrator would include the “phantom account” in his calculations. The total value of this phantom account in 1989, when John rejoined Xerox, is equal to John’s lump-sum distribution of $140,000 x 1.059, or approximately $217,200.)
The Plan Administrator would thus compare John’s pension, column D, and column H to determine John’s benefit. As you can see above, column H provides the greatest benefit, so John’s cash account would be used to calculate the benefits he would receive upon retirement.
Second, the Plan Administrator would “offset” John’s prior distribution against his current benefits to determine the amount of benefits John would actually receive. Thus, the Plan Administrator would take the “total” value of John’s cash account, including the “phantom account” ($277,800), and subtract out the value of the “phantom account” ($264,000). The total present value of the benefits John would receive upon his second retirement would thus be $13,800.
This means that John would receive approximately $690 annually ($13,800 x 5 percent) upon retirement under the Plan Administrator’s “phantom account” approach. In comparison, if John had simply been treated as a new employee when he was rehired, his pension would have entitled him to at least $3,500 annually ($50,000 x 1.4 percent x 5 years) upon his retirement. And the impact of the “phantom account” *541may have been even more dramatic with respect to some of the respondents in this case. See Brief for Respondents 24 (describing how respondent Paul Frommert erroneously received a report claiming that his retirement benefits were $2,482.00 per month, before later discovering that, because of the “phantom account,” his actual monthly pension was $5.31 per month); see also App. 63a.