Joy Rogers Thomas was fired by Metroflight, Inc. after she married a fellow employee, because of a Metroflight “no-spouse” employment rule. Thomas sued Metroflight for employment discrimination under Title VII of the Civil Rights Act of 1964, 42 U.S.C. §§ 2000e to 2000e-17. After a bench trial the district court entered judgment for Metroflight and ruled that Metroflight was entitled to attorney fees. The appeal was submitted on the briefs by agreement of the parties.
Metroflight is a small commercial airline doing business in Oklahoma and employing about 500 people. Metroflight hired Thomas as a secretary for its Lawton, Oklahoma *1508operations in 1981, and in 1983 Thomas married a Metroflight pilot. At the time of her marriage, Thomas’ duties were seventy-five percent in the company’s maintenance department and twenty-five percent in its flight operations department, where Thomas’ husband worked.
There was a “no-spouse” employment rule in effect at Metroflight at the time of Thomas’ marriage, under which no two persons in the same department could marry and then continue working for Metroflight. Any two such persons who did marry had the option of choosing which spouse would quit. If neither quit, the company would fire the employee with the lesser seniority. Neither Thomas nor her husband quit, so Thomas, who had less seniority than her spouse, was fired.
The record shows that before Thomas’ firing there had been eight other instances of intrafirm marriage at Metroflight. In seven of the previous instances, either the no-spouse rule was not violated because the spouses worked in different departments, or accommodations were made to retain both employees by reclassifying one spouse’s work assignment or simply allowing the violation. In one other instance the rule was enforced, also by firing the female employee.
After being fired, Thomas filed a discrimination complaint with the Equal Employment Opportunity Commission (EEOC), which found no violation but issued Thomas a “right-to-sue” letter. Thomas then initiated this Title VII suit in federal district court, relying at trial solely on a theory of the disparate impact of Metroflight’s no-spouse rule.1
After hearing the evidence, the district court found that Metroflight’s no-spouse policy “is a good policy, a proper company business decision and is not discriminatory.” R. I, 204; R. XI, 121. The court held that Thomas did not establish any disparate impact of the policy, finding her “statistical evidence to be wholly inadequate and irrelevant in that the fact that salary differentials may be considered, among other factors, by a couple in deciding which will resign under the defendant’s rule is not a sufficient ground for any legally cognizable claim.” R. I, 204. The court rejected Thomas’ statistical analysis because of “gross overstatement of base data,” “small actual sample size,” “absence of any causal connection between spousal consideration of salary” and termination by Metroflight based on seniority, “failure to measure actual adverse impact rather than a purely conceptual potential adverse impact,” and “total disregard of the relevant, real world factors.” Id. at 205. At the same time, the court directed defendant Metroflight to file a motion for attorney’s fees with supporting affidavits, stating from the bench that there was “little reason or excuse, in my judgment, for having fooled with this case in the first place.” R. XI, 122. We construe this as a finding that plaintiff Thomas’ action was frivolous.
I
In Title VII cases based upon alleged discriminatory intent — disparate treatment — the plaintiff must present evidence giving rise to “an inference of unlawful discrimination.” Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 253-54, 101 S.Ct. 1089, 1093-94, 67 L.Ed.2d 207 (1981). When the plaintiff has done that, the employer need only present a plausible nondiscriminatory explanation for its actions to dispell the otherwise legally mandatory inference. The plaintiff then has an opportunity to show that the employer’s proffered explanation is pretextual; the burden of proof never shifts from the plaintiff. Id. at 254-56, 101 S.Ct. at 1094-95.
Proof of discriminatory intent is not necessary for liability under Title VII, however. In Dothard v. Rawlinson, 433 U.S. 321, 97 S.Ct. 2720, 53 L.Ed.2d 786 (1977), the Supreme Court expressly held that disparate impact alone, without a showing of discriminatory intent, is sufficient to make out a prima facie case of sexual discrimina*1509tion unlawful under Title VII. Dealing in that case with hiring standards, the Court stated that “to establish a prima facie case of discrimination, a plaintiff need only show that the facially neutral standards in question select applicants for hire in a significantly discriminatory pattern.” Id. at 329, 97 S.Ct. at 2726. The same reasoning applies to a rule requiring termination of one spouse when a marriage occurs between employees of a company.
In a disparate impact case, unlike a disparate treatment case, once a plaintiff makes out the prima facie case, the burden of proof shifts to the employer to show “business necessity:”
“In a disparate impact case, ... the employer must prove business necessity for the challenged practice to rebut the prima facie case. He bears a burden of proof. Moreover, in a disparate impact case, unlike a disparate treatment case, a rational or legitimate, nondiscriminatory reason is insufficient. The practice must be essential, the purpose compelling.”
Williams v. Colorado Springs, Colorado School District, 641 F.2d 835, 842 (10th Cir.1981) (emphasis in original) (citations omitted); see also Hawkins v. Bounds, 752 F.2d 500, 503-04 (10th Cir.1985); Lasso v. Woodmen of the World Life Insurance Co., 741 F.2d 1241, 1245 (10th Cir.1984), cert. denied, 471 U.S. 1099, 105 S.Ct. 2320, 85 L.Ed.2d 839 (1985). This is, of course, a heavier burden upon the employer than in discriminatory treatment cases. The different allocations of the burdens of persuasion and production in disparate treatment and disparate impact cases stem from the different requirements for establishing the prima facie case:
“ '[Establishing a prima facie case of disparate treatment is not onerous.’ Burdine, 450 U.S. at 253, 101 S.Ct. at 1094. In making a prima facie case in a disparate impact suit, however, the plaintiff must not merely prove circumstances raising an inference of discriminatory impact; he must prove the discriminatory impact at issue.”
Johnson v. Uncle Ben’s, Inc., 657 F.2d 750, 753 (5th Cir.1981), cert. denied, 459 U.S. 967, 103 S.Ct. 293, 74 L.Ed.2d 277 (1982) (emphasis added); accord Segar v. Smith, 738 F.2d 1249, 1266-68, 1270 n. 17 (D.C.Cir.1984), cert. denied, 471 U.S. 1115, 105 S.Ct. 2357, 86 L.Ed.2d 258 (1985); Moore v. Hughes Helicopters, Inc., 708 F.2d 475, 481-82 (9th Cir.1983).
We emphasize the differences between the two types of cases because we agree with the district court that Thomas did not produce sufficient evidence to establish the discriminatory impact of the no-spouse rule. We do not agree, however, with all that the district court said in arriving at its conclusion. Indeed, we affirm on this issue reluctantly because we suspect, as others have claimed,2 that “no-spouse” rules in practice often result in discrimination against women, and are generally unjustified. But we cannot accept our own speculations or others’ conclusions as a substitute for plaintiff’s required proof, nor may we take judicial notice of evidence that might have been but was not presented.
Thomas presented evidence that in the only two instances in which Metroflight enforced the “no-spouse” rule, female employees were fired. But that alone is insufficient to prove a violation of Title VII; a sample of two is too small to make even a 100% impact rate significant.3 Compare Harper v. Trans World Airlines, 525 F.2d 409 (8th Cir.1975) (evidence that four of five applications of similar no-spouse rule resulted in female job losses held inadequate without more to show disparate impact), with Yuhas v. Libby-Owens-Ford *1510Co., 562 F.2d 496 (7th Cir.1977) (“substantial discriminatory impact” established when seventy-one of seventy-four applicants excluded by no-spouse hire rule were women), cert. denied, 435 U.S. 434, 98 S.Ct. 1510, 55 L.Ed.2d 531 (1978) and EEOC Dec. 75-239, 2 Empl.Prac.Guide (CCH) ¶ 6492, at 4260-61 (Mar. 2, 1976) (disparate impact established when sixty-five of sixty-six applicants rejected because of no-spouse rule were female). See also Morita v. Southern California Permanente Medical Group, 541 F.2d 217, 220 (9th Cir.1976) (sample of eight too small), cert. denied, 429 U.S. 1050, 97 S.Ct. 761, 50 L.Ed.2d 765 (1977); Robinson v. City of Dallas, 514 F.2d 1271, 1273 (5th Cir.1975) (sample of twelve too small). But see Bunch v. Bullard, 795 F.2d 384, 395 (5th Cir.1986) (disparate impact established in small sample of twenty-eight, where observed disparity in exam pass rates was great enough); Marsh v. Eaton Corp. 639 F.2d 328, 329 (6th Cir.1981) (error for trial court to conclude that disparate impact could not be established, after evaluation of small pool of forty-four new hires revealed eight men placed in slightly better positions than women; noting danger of simply exempting small employers from Title VII); accord Pegues v. Mississippi State Employment Service, 699 F.2d 760, 769 (5th Cir.), cert. denied, 464 U.S. 991, 104 S.Ct. 482, 78 L.Ed.2d 679 (1983).
Thomas also presented statistical evidence to demonstrate the disparate impact of the no-spouse rule, through the testimony of a statistical expert, James Horrell. Using employment data provided by Metroflight, Horrell testified that considering all possible marriages between two employees in the same department of Metroflight, if the decision who would terminate were always made on the basis of preserving the higher salary, more women than men would terminate. To come to this conclusion the expert assumed that any employee in a Metroflight department might marry any other opposite-sex employee in the department, regardless of either employee’s age or current marital status. From a universe of 3687 possible marriages thus calculated, he found that in 62.1% of the marriages the woman would have the lesser salary. For a subset of 1283 possible marriages in which the annual salary differential would be more than $5000, the percentage of women receiving less, and thus terminating, would be 85.4%. PL Ex. 3.
The Metroflight no-spouse rule, of course, does not require that the spouse who receives the lower pay terminate. Horrell's assumption that salary differential would be the deciding factor in a couple’s decision is thus critical to Thomas’ case. But Horrell did not testify as an expert on spousal decision-making. On cross-examination, he admitted that other factors, such as availability of positions with other employers, could be a significant element in a married couple’s decision as to which should terminate. He stated, however, that he did not have information on such factors and would find them nearly impossible to quantify statistically.
Horrell also presented calculations— based upon the same assumption that every departmental employee was a potential spouse of every other opposite-sex departmental employee — that, if seniority were to be the sole basis of the decision, the woman would terminate in 52.4% of the cases. The no-spouse rule, of course, mandates that seniority will determine which spouse’s employment is terminated only if the couple does not make the decision. Thomas introduced no other evidence on what would affect a couple’s decision as to which spouse would terminate employment, except for her own somewhat ambiguous testimony: “We did discuss it somewhat at length, of course, I didn’t want to give up my job and because of economic reasons, the difference in the pay, there is really no decision as to who was going to be terminated, it definitely was going to be me, I was the most junior.” R. XI, 13.
The evidence Thomas presented is sufficient to prove disparate impact if salary is the controlling factor, and perhaps sufficient to prove disparate impact if seniority is the controlling factor.4 But *1511Thomas presentation was insufficient to prove that the no-spouse rule as adopted by Metroflight has a disparate impact on women because Thomas presented no evidence that salary or seniority would in fact control the decision as to which spouse would terminate employment. Thomas presented only assumptions to that effect. We hold that is not enough.
We do not believe that the presence of choice is enough to make discriminatory impact unprovable. Many Title VII decisions involve employee choices. See, e.g., International Brotherhood of Teamsters v. United States, 431. U.S. 324, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977) (job rule requiring loss of seniority, enforced only after employee choice to switch driving routes, presents potential Title VII disparate impact and disparate treatment violation); cf. Bhatia v. Chevron U.S.A., Inc., 734 F.2d 1382, 1384 (9th Cir.1984) (job rule requiring termination for refusal to shave enforced only after employee choice to keep beard style required by Sikh religion presents potential Title VII violation). The plaintiff need not prove that only one factor, such as money, controls the decision in all cases. What the plaintiff must prove by statistical studies, evidence of experience in sufficiently similar circumstances, expert testimony on decision-making in the context of no-spouse rules, or other evidence, is that the choice made will result in the termination of the woman’s employment in significantly more instances. We cannot supply the proof that is missing here on the basis of our own or others’ assumptions without evidence relevant to the case before us.
Proof of how the no-spouse rule affects the actual choices a couple makes may be difficult. This is acknowledged by commentators, including some who have suggested amendments to Title VII to eliminate the problem.5 Whether it is impossible to prove or merely difficult, we cannot waive the requirement that the plaintiff show the disparate impact. Thomas’ evidence in this case was insufficient to do that. Therefore, we agree with the district court that Thomas did not establish the *1512disparate impact of Metroflight’s no-spouse rule.
II
The district court, sua sponte, found that plaintiff Thomas’ case was frivolous, and it directed Metroflight to make an application for attorney’s fees to be paid by Thomas.
On an issue on which the circuits are split, this court adopted a rule in December 1986 that a district court’s order awarding attorney’s fees is final for purposes of appeal only after the amount is determined. Phelps v. Washburn University, 807 F.2d 153 (10th Cir.1986). In Phelps the district court’s order holding that defendants were entitled to attorney’s fees was entered separately from and several months after its order granting summary judgment to defendants and after plaintiffs had filed an appeal on the merits of their case. We held that a separate appeal of the attorney’s fees ruling, before the court set the amount, was premature.
In the instant case, the judge determined Thomas’ liability for Metroflight’s attorney’s fees, though not the amount, at the same time he announced his findings and conclusions on the merits of the suit. Thomas’ brief on appeal challenged the frivolousness finding, as a part of her appeal on the merits. We understand that the district court set the amount of the fees at something over $30,000 sometime in 1985, and that Thomas filed no new appeal within the permitted time limits, which expired long before we entered our Phelps opinion.
Phelps established a new rule for this circuit, imposing a requirement on Thomas that she had no opportunity to meet in timely fashion. Retroactive application of the rule would impose a substantial inequity on her. See EEOC v. Gaddis, 733 F.2d 1373, 1376 (10th Cir.1984). Therefore, we will consider here whether plaintiff’s case was frivolous.
Thomas’ challenge to a no-spouse rule as a discriminatory employment practice presents an issue of first impression in this circuit. It is error to assess fees against a civil rights plaintiff when a non-frivolous issue is one of first impression at the circuit level. See Kiowa Tribe v. Lewis, 777 F.2d 587 (10th Cir.1985), cert. denied,—U.S.-, 107 S.Ct. 247, 93 L.Ed.2d 171 (1986). Moreover, no-spouse employment rules have spawned difficult cases in other courts. See, e.g., EEOC v. Rath Packing Co., 787 F.2d 318 (8th Cir.), cert. denied,—U.S.-, 107 S.Ct. 307, 93 L.Ed.2d 282 (1986); George v. Farmers Electric Cooperative, Inc., 715 F.2d 175 (5th Cir.1983); Yuhas v. Libbey-OwensFord Co., 562 F.2d 496 (7th Cir.1977), cert. denied, 435 U.S. 934, 98 S.Ct. 1510, 55 L.Ed.2d 531 (1978); Harper v. Trans World Airlines, 525 F.2d 409 (8th Cir.1975). Plaintiff Thomas failed to prove disparate impact in this case, but her attempt cannot be characterized as frivolous or illusory.
We therefore reverse the district court’s ruling that the case was frivolous and its order directing plaintiff to pay defendant’s attorney’s fees.
AFFIRMED IN PART, REVERSED IN PART, and REMANDED for further proceedings consistent herewith.