This is a petition by Equifax Inc. to review a portion of a Final Order to Cease and Desist issued by the Federal Trade Commission on December 15, 1980.1 The Commission found that Equifax was guilty of six major violations of the Fair Credit Reporting Act (FCRA), 15 U.S.C. §§ 1681-1681t. Equifax challenges only one of the Commission’s findings: that Equifax failed to use reasonable procedures to assure maximum possible accuracy in its consumer reports in violation of section 607(b) of the FCRA, 15 U.S.C. § 1681e(b).2
I. STATUTORY BACKGROUND
The stated purpose of the FCRA is “to prevent consumers from being unjustly damaged because of inaccurate or arbitrary information in a credit report.” 3 Congress *1049apparently recognized, however, that total accuracy in consumer reports is not a realistic objective. Accordingly, section 607(b) of the FCRA provides:
Whenever a consumer reporting agency prepares a consumer report it shall follow reasonable procedures to assure maximum possible accuracy of the information concerning the individual about whom the report relates.
15 U.S.C. § 1681e(b).4
FCRA provides for two methods of enforcement: private damage suits brought under sections 616 and 617, 15 U.S.C. §§ 1681n, 1681o, and administrative actions under section 621, 15 U.S.C. § 1681s. The present proceeding was brought before the Commission under the provisions of section 621(a).5
Administrative enforcement of FCRA is committed to eight other agencies in addition to the Commission.6 In the discharge of its responsibility the Commission acts in an enforcement role only. Congress has not entrusted the Commission with rule making authority under FCRA.7
II. THE ENFORCEMENT PROCEEDINGS
Equifax, which was formerly known as Retail Credit Co., is one of the nation’s largest consumer reporting agencies. It has over a thousand offices and sub-offices. During the relevant period it had approximately 4,600 salaried field representatives *1050and approximately 17,000 customers.8 It provides over twenty million consumer reports annually. The bulk of those reports concerns applicants for life, health, fire, property and automobile insurance.
The complaint in this case was issued on February 21, 1974 after a nationwide investigation that had lasted over two years. The alleged violations involved in the present petition center around certain aspects of the Equifax quality control program. Specifically, Equifax sampled consumer reports from its branch offices, tabulated the amounts of adverse information and used the tabulation to rank branch offices into upper, middle and lower third positions.
As defined by the Commission adverse information means,
information which may have, or may reasonably be expected to have, an unfavorable bearing on a consumer’s eligibility or qualifications for credit, insurance, employment, or other benefit, including information which may result, or which may be reasonably expected to result, in a denial of or increased costs for such benefits.9
The term does not necessarily refer to information that would reflect on the character or morals of an applicant or that would ordinarily be regarded as derogatory. It could, for example, include lack of experience for a job applicant, the presence of youthful drivers in the family of an automobile insurance applicant, a dangerous hobby such as parachuting or scuba diving for a life insurance applicant or obvious overweight of a health insurance applicant.
The FTC alleged that the tabulated rankings were utilized to apply pressure on Equifax field offices to increase their production of adverse information, and that this pressure created an unreasonable risk that Equifax employees would fabricate such information. Evidentiary hearings began on June 30, 1975 and the record was closed on June 15, 1977. Three hundred seventeen witnesses, including two hundred twenty-one present and former Equifax employees, testified and several thousand pages of exhibits were received. The Administrative Law Judge (hereinafter “ALJ”) made his initial decision on November 11, 1977, concluding that Equifax violated section 607(b). The findings of the ALJ were summarized by him in part as follows:
The record does not demonstrate an overall pattern of overt sanctions such as firings or the withholding of salary increases or promotions as penalties for the failure to achieve specific levels of [adverse] information. The pressures as far as can be determined from this record, were generally more subtle: for example, achievement in production of [adverse] information as measured by the audits was one factor considered in connection with managerial incentive bonuses and field representatives put down the objective of achieving particular percentages of [adverse] information on their personal appraisal forms. And, field representatives’ performance on the quality audits was one factor among several to be considered in salary recommendations by the branch manager.
Initial Decision of ALJ, p. 238 (footnotes omitted).
The ALJ also recognized, however, that Equifax’s quality control is not limited to the tabulation of adverse information. It also involves the selection and training of its field representatives, and their instructions requiring scrupulous honesty and fairness. Equifax conducts periodic quality audits of each branch office to check performance and the integrity of their reports. In addition to adverse information they also tabulate below standard underwriting (defects in a report’s content or completeness) and below standard operating (clerical defects) and offices are also ranked with respect to these deficiencies. The ALJ found that it is Equifax’s policy to terminate immediately any representative falsifying information, that this policy is communicated *1051to the field and is generally carried out. Although the ALJ made detailed findings as to ways in which subtle pressure resulted from Equifax’s use of its tabulation of adverse information production, he conceded “there is no evidence in the record of a report where adverse information has been falsified.” Initial Decision of ALJ, p. 243.
The Commission’s decision and final order was issued on December 15,1980. It rejected certain findings of the ALJ, limited the scope of his order as to others, broadened his order with respect to the questions now before this court and otherwise adopted his decision, findings and conclusions. Notwithstanding the absence of any proof of inaccuracy by Equifax in the production of adverse information, the Commission held:
A procedure which encourages the production of adverse information is likely to lead employees to prepare their reports in a manner detrimental to the legitimate interests of the consumers about whom reports are written. This risk was realized in this case. Because we can hardly conclude that procedures which pressure employees to produce adverse information are necessary to the proper operation of consumer reporting agencies, and are unaware of any justification for their use which would outweigh the risks which they pose, we find that they violate Section 607(b).10
(footnotes omitted).
III. REVIEW OF THE FTC ORDER
The Commission’s holding and the provisions of the order based upon it are attacked by Equifax on three grounds: (1) that there cannot be a section 607(b) violation without proof of inaccuracy, (2) that as interpreted by the Commission section 607(b) is an unconstitutional prior restraint on commercial speech, and (3) that there is no substantial evidence to support the Commission’s finding that the Equifax procedure violated the section. Because of our disposition of the case, we find it unnecessary to consider the constitutionality of the Commission’s order.
Equifax stoutly maintains that to establish a section 607(b) violation it is essential that inaccuracy be proved. It asserts that every court that has considered the section has held that proof of inaccuracy is the first requisite to proving a violation. See Bryant v. TRW, Inc., 487 F.Supp. 1234 (E.D. Mich.1980); McPhee v. Chilton Corp., 468 F.Supp. 494 (D.Conn.1978); Todd v. Associated Credit Bureau Services, 451 F.Supp. 447 (E.D.Pa.1977), aff’d mem., 578 F.2d 1376 (3d Cir. 1978), cert. denied, 439 U.S. 1068, 99 S.Ct. 834, 59 L.Ed.2d 33 (1979); Lowry v. Credit Bureau, Inc., 444 F.Supp. 541 (N.D.Ga.1978); Roseman v. Retail Credit Co., 428 F.Supp. 643 (E.D.Pa.1977); Pendleton v. Trans Union Systems Corp., 76 F.R.D. 192 (E.D.Pa.1977); Middlebrooks v. Retail Credit Co., 416 F.Supp. 1013 (N.D.Ga. 1976); Austin v. Bankamerica Service Corp., 419 F.Supp. 730 (N.D.Ga.1974); Feller v. Retail Credit Co., 359 F.Supp. 1235 (N.D.Ga.1973), aff’d mem., 505 F.2d 733 (5th Cir. 1974).
The Commission responds that these cases were brought by private plaintiffs under FCRA sections 616 and 617 and are inappo-site to an enforcement action initiated by the Commission under section 621. It argues that whereas a private plaintiff necessarily must prove injury to recover damages, the Commission may act to enjoin a practice posing a risk of harm to consumers before that harm actually occurs. It says that the meaning and legislative history of the section establish the Commission’s prophylactic enforcement role.
For purposes of our initial analysis we assume without deciding that the position of the Commission is correct. In view of our conclusion it is not necessary that we resolve the question.
Our review of the Commission’s finding that Equifax violated section 607(b) is narrow in scope. The findings of the Commission are conclusive if they are supported by substantial evidence. Additionally, reasonable inferences drawn from facts based upon substantial evidence may sup*1052port the findings. See Colonial Stores, Inc. v. FTC, 450 F.2d 733, 739 & n.13 (5th Cir. 1971). The fact that two reasonable inferences could be drawn from the evidence does not detract from the Commission’s decision to choose one of those inferences. U. S. Pipe & Foundry Co. v. Webb, 595 F.2d 264, 266 (5th Cir. 1979); Keele Hair & Scalp Specialists, Inc. v. FTC, 275 F.2d 18, 21 (5th Cir. 1960). We must take into account, however, evidence that fairly detracts from the Commission’s findings. Universal Camera Corp. v. NLRB, 340 U.S. 474, 488, 71 S.Ct. 456, 464, 95 L.Ed. 456 (1951); Abilene Sheet Metal, Inc. v. NLRB, 619 F.2d 332, 337 (5th Cir. 1980).
The posture of the issues in this case is somewhat unusual. The controlling facts are not in serious dispute.11 The evidence amply supports the Commission’s factual findings that Equifax maintains a statistical tabulation of the production of adverse information by its offices, that its methodology is statistically flawed and that it uses such tabulation as a subtle device to encourage production of adverse information. The ALJ and the Commission also found, however, that Equifax strictly enforced a policy against the inaccurate production of adverse information, and that the Commission’s survey of Equifax reports revealed that the inaccuracies, if anything, favored the consumer. The question is whether this constitutes substantial evidence of a violation of section 607(b). Phrased another way, do the facts found by the Commission and based upon substantial evidence support a reasonable inference that the quality control system at Equifax is likely to produce error? We conclude that the facts do not support such an inference, and we therefore set aside the challenged portions of the FTC order.
The essential task of the FTC in determining whether a credit reporting agency violated section 607(b) is to ascertain whether the procedures followed by the agency pose an unreasonable risk of producing error. We have specifically reserved the question whether the agency must show that the risk of error manifested itself in concrete instances of detriment to the credit consumer. In this case, however, the FTC attempted mightily to discover whether the quality control procedure resulted in actual inaccuracies. This attempt formed part of an enormous investigation, national in scope, extending over two years with hundreds of witnesses and thousands of pages of documents. The investigation failed to produce any substantial evidence of inaccuracies in the production of adverse information.12 It is not reasonable to conclude that a procedure which produces no evidence of inaccuracy despite the intensive scrutiny of the FTC over a several year period is a procedure that poses an unreasonable risk of inaccuracy.
The analysis of the court in Firestone Tire & Rubber Co. v. FTC, 481 F.2d 246 (6th Cir.), cert. denied, 414 U.S. 1112, 94 S.Ct. 841, 38 L.Ed.2d 739 (1973), is instructive for our purposes. In that case, the FTC found that the petitioner had engaged in deceptive advertisement. Part of the evidence was a consumer survey showing that 15% of a scientifically selected sample of consumers were misled by the ad in question. The court found that the Commission could reasonably infer under these circumstances that the ad was deceptive.13 The court *1053cautioned, however, that if the evidence had affirmatively shown that consumers were not misled, the court could not routinely accept the Commission’s inference. Id. at 249. See also Boise Cascade Corp. v. FTC, 637 F.2d 573, 578-80 (9th Cir. 1980) (theory regarding likely effect of pricing policy was not substantial evidence of anticompetitive practice where, inter alia, lack of buyer objection to practice indicated that policy did not have anticompetitive effect); American Optometric Ass’n v. FTC, 626 F.2d 896, 912-13 (D.C.Cir.1980) (speculation regarding probable regulatory actions by the States was insufficient basis on which to predicate findings when actual record of state actions might disprove speculation); Ger-Ro-Mar, Inc. v. FTC, 518 F.2d 33, 36-38 (2d Cir. 1975) (mathematical extrapolation performed by Commission indicating that marketing plan contained improbable projections was not substantial evidence of deceptive practice where, inter alia, Commission’s witness testified he was not actually deceived).14
It is also noteworthy that the tabulation of adverse information was merely one part of an overall program of quality control which involved personnel selection, training, supervision, review and various other tabulations. There is nothing aberrant about Equifax’s concern that adverse information be reported. Its customers want accurate information, to be sure, but it is self evident that their central concern is that they also receive complete information including all adverse information that exists. The quality control procedure also includes unequivocal requirements for fairness and a stringent prohibition against falsification of adverse information, found by the ALJ to be enforced by Equifax. Cf. Marcus v. FTC, 354 F.2d 85, 89 (2d Cir. 1965) (scant evidence of mislabeling was insufficient basis on which to predicate findings in light of unrefuted testimony that great care was taken to assure accurate labeling).
Under the peculiar circumstances of this case, the evidentiary basis is insufficient to support a rational inference that Equifax’s quality control audit procedures, including the tabulation of adverse information and Equifax’s challenged use of that tabulation, would likely result in inaccurate information. Such an inference can hardly be judged reasonable when it flies squarely in the face of the Commission’s own findings.
We hold that Paragraphs C and D of Part I of the Commission’s Final Order, the only portion before us for review, is not based upon reasonable inferences drawn from the record evidence and is set aside.
ORDER SET ASIDE IN PART.