In this appeal, we consider the circumstances in which defendants may recover attorney’s fees from plaintiffs in actions under Title VII.
In 1984, the American Federation of State, County and Municipal Employees, one of its affiliates and several individual employees (collectively referred to as “AFSCME”) commenced this civil rights action against Nassau County and several of its officers (variously, the “County” and the “defendants”). AFSCME alleged that the County’s job classification system assigned lower salary grades to jobs that were predominantly held by women than to jobs that were predominantly held by men, in violation of Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq., and the Equal Pay Act of 1963, 29 U.S.C. § 206(d). In 1985, the United States District Court for the Eastern District of New York (Glasser, J.) dismissed AFSCME’s disparate impact claim, but declined to dismiss AFSCME’s disparate treatment claim. AFSCME v. County of Nassau, 609 F.Supp. 695, 708-12 (E.D.N.Y.1985). Under the remaining disparate treatment theory, AFSCME was required to establish that the County intended to discriminate against women in implementing the job classification system. See International Brotherhood of Teamsters v. United States, 431 U.S. 324, 335 n. 15, 97 S.Ct. 1843, 1854-55 n. 15, 52 L.Ed.2d 396 (1977).
On August 24, 1992, following a 16-day bench trial, Judge Glasser issued a memorandum and order holding, among other things, (i) that AFSCME failed to establish discriminatory intent with respect to its claim that the County’s implementation of its pay system discriminated against all women who worked in female-dominated jobs; but (ii) that AFSCME did succeed in establishing that the County intentionally discriminated in paying police detention aides (a position predominantly held by women) less than turnkeys (a better-paid position predominantly held by males), in violation of Title VII. AFSCME v. County of Nassau, 799 F.Supp. 1370, 1414, 1416 (E.D.N.Y.1992). The district court ordered the County to pay the female police detention aides a total of $1.6 million in back pay for a fourteen-year period. No appeal was taken from that decision.
The County moved for an award of attorney’s fees and expert witness fees pursuant to 42 U.S.C. § 2000e-5(k), which provides for recovery of such fees by prevailing parties. In 1993, the district court held that AFSCME’s claims were frivolous under Christiansburg Garment Co. v. EEOC, 434 U.S. 412, 98 S.Ct. 694, 54 L.Ed.2d 648 (1978), and awarded the defendants $982,407.23 in attorney’s fees. AFSCME v. County of Nassau, 825 F.Supp. 468, 473-74 (E.D.N.Y.1993). The Court deferred decision on the defendants’ motion for expert witness fees pending the outcome of a case that was then before the Supreme Court concerning the retroac-tivity of a 1991 amendment to 42 U.S.C. § 2000e-5(k) that expressly made expert witness fees available to prevailing parties. Id. *646at 474. In an unpublished memorandum and order dated May 31, 1995, the district court held that the 1991 amendment applied retroactively under Landgraf v. USI Film Products, 511 U.S. 244, 114 S.Ct. 1483, 128 L.Ed.2d 229 (1994), and therefore awarded the County $550,974.66 in expert witness fees. AFSCME appeals both of those decisions.
Attorney’s fees and expert witness fees may not be awarded to a prevailing defendant in a Title VII case unless the plaintiffs claim is “frivolous, unreasonable, or groundless, or ... the plaintiff continued to litigate after it clearly became so.” Christiansburg, 434 U.S. at 422, 98 S.Ct. at 701; see also Rounseville v. Zahl, 13 F.3d 625, 632 (2d Cir.1994). AFSCME established a prima fa-cie case of a violation of Title VII; introduced extensive statistical evidence through the testimony of two expert witnesses which, if credited, could have supported a judgment in its favor; and won $1.6 million on its claim that the County intentionally discriminated in paying police detention aides (predominantly women) less than turnkeys (predominantly men). AFSCME, 799 F.Supp. at 1416. Under these circumstances, we conclude that AFSCME’s claims were not “frivolous, unreasonable or groundless.” We therefore hold that the district court abused its discretion in awarding attorney’s and expert witness fees to the County, and reverse the judgments.
BACKGROUND
Nassau County, New York, employs more than 19,000 people in more than 1,500 jobs. Their salaries are determined by the salary grade assigned to their job title and their seniority. Id. at 1373-74. The present compensation system was implemented after a comprehensive job and salary evaluation was conducted between 1964 and 1967. The details of the job evaluation process (called the “Cresap” process after the management consulting firm that assisted the County) are lucidly explained in the district court’s 1992 opinion, see id. at 1374-77, and will only be summarized briefly herein.
First, a team of management consultants and County managers surveyed several thousand employees. After assessing the skills required and tasks performed in each occupation, they created a taxonomy of 650 job titles. Id. at 1375. The team then composed job specifications that described the characteristics of each job in terms of certain prescribed criteria (i.e., the typical duties, the complexity of the duties, and the required qualifications). The County assigned a numerical salary grade to four factors for each job title: (i) the skills and knowledge required by the job; (ii) the variety and complexity of the job duties; (iii) any responsibility for independent action; and (iv) any supervisory responsibility. Id. The numerical values for the four factors for each job title were then weighted and totaled. Other information gathered during the classification process was used to revise the totals. Id. at 1376. Finally, the scores were organized into twenty-seven ranges to correspond with salary grades ranging from one to twenty-seven. Id. (It appears that no consideration was given to the wage that any job would command in the labor market.) The County passed an ordinance enacting the final job titles and salary grades in May 1967. Id. at 1377.
As of the time of trial, three-quarters of the County’s employees held job titles established by the Cresap process, and were paid in accordance with the salary grades adopted in 1967.
Between 1967 and 1991, the job classification system was altered by the County in three ways: (i) several new job titles were created by the County’s Civil Service Commission; (ii) the pay grades of several job titles were raised by the Civil Service Commission; and (iii) higher pay grades and salary levels were assigned to several job titles as a result of collective bargaining. Id. at 1377-78.
AFSCME is the collective bargaining representative for the employees of Nassau County. AFSCME’s amended complaint alleged that the County “discriminates in compensation on the basis of sex by paying historically female job classifications less than historically male classifications which require an equivalent or lesser composite of skill, *647effort, responsibility and working conditions,” in violation of Title VII and the Equal Pay Act. Id. at 1372 (quoting amended complaint, ¶ 3(D)). The amended complaint asserted rights on behalf of all employees of the County, and defined “historically female” and “historically male” job classifications as those in which more than 70% of the positions have always been held by persons of that sex. Id. The individual plaintiffs alleged specific violations of Title VII and the Equal Pay Act with respect to their particular job titles.
Prior to discovery, the County moved to dismiss all of AFSCME’s claims. On May 17, 1985, the district court dismissed the Equal Pay Act claims of three of the individual plaintiffs, and dismissed the Title VII claims of all of the plaintiffs “insofar as they seek relief under [a] disparate impact theory.” AFSCME, 609 F.Supp. at 713. But the court denied the County’s motion to dismiss with respect to AFSCME’s disparate treatment claim, and granted AFSCME leave to amend its complaint to better plead a disparate impact claim. Id. AFSCME never made such an amendment. On July 13,1987, the court dismissed the Title VII claims of all the County’s male employees for lack of standing; certified a class composed of all the female employees who, since July 28, 1982, worked in jobs held at least 70% by women; and designated two individual employees and AFSCME and its affiliate, Local 1000 of the Civil Service Employees Association, to represent the class. AFSCME v. County of Nassau 664 F.Supp. 64, 67, 69 (E.D.N.Y.1987). Extensive discovery took place from 1985 to 1989.
At trial, AFSCME proceeded under two theories: that the County intentionally discriminated against all women in all historically female positions, and that the County intentionally discriminated against particular women in three particular historically female positions.
As to its broader claim, AFSCME contended that during the Cresap process the County “ ‘selectively departed from [its wage setting] methodology in a deliberate effort to favor male-dominated job titles while disfavoring the most significant female-dominated titles,”’ in violation of Title VII and the Equal Pay Act. AFSCME, 799 F.Supp. at 1378 (quoting AFSCME’s post-trial brief). Because AFSCME never amended its complaint to adequately plead a disparate impact claim, AFSCME was required to introduce evidence from which it could be inferred that the County intentionally discriminated in setting the pay scale in order to prevail under a disparate treatment theory. See Teamsters, 431 U.S. at 335 n. 15, 97 S.Ct. at 1854-55 n. 15 (discriminatory intent must be shown in disparate treatment cases).
AFSCME also contended that the County violated the Equal Pay Act and Title VII by setting lower salaries for three particular job groups held predominantly by women. In particular, AFSCME contended that police communication operators, mainly women, were paid less than fire communications technicians, mainly men; that police detention aides were paid less than turnkeys; and that clerical workers were paid less than correction officers assigned to clerical tasks. AFSCME, 799 F.Supp. at 1407-1409.
At trial, AFSCME presented no direct evidence that the County intentionally discriminated in setting pay, but instead relied on statistical evidence tending to show that jobs filled predominately by women were paid less than jobs filled predominately by men. See Teamsters, 431 U.S. at 339-40, 97 S.Ct. at 1856-57 (Title VII plaintiff may rely on statistics in a disparate treatment case to show discriminatory intent). AFSCME relied on the testimony of two expert witnesses: Dr. Stephan Michelson, a labor economist, and Dr. Donald J. Treiman, a sociologist. Dr. Michelson’s testimony was offered to show that the County currently discriminated by failing to correct disparities in pay, and by employing methods to assign pay grades that resulted in disparities. AFSCME, 799 F.Supp. at 1395. He testified about a study he made of the County’s current job classification and wage system in which he concluded that, after more than 30 gender-neutral variables were accounted for, there remained a gap of approximately $3000 between equivalent historically male and historically female jobs. Id. at 1396. Michelson averred that *648this gap could be explained only by the sex of the County’s employees. Id.
Dr. Treiman’s testimony was offered to show that the County intentionally discriminated in establishing salary grades during the 1967 Cresap process. Dr. Treiman testified about two studies. The first was his study of the County’s job classification system, in which he attempted to replicate the 1967 Cresap process in order to see if he would reach a similar result. Id. at 1379. Five graduate students were hired to rate each job title in accordance with the four factors used by the County in the Cresap process, and to assign each job to one of twenty-seven pay grades in accordance with those values. Dr. Treiman then compared the results of this study with the actual pay grades and job classifications used by the County. The sociologist concluded that the County had “not applied the job evaluative criteria in a ‘good faith’ manner,” and that the County “ ‘applied the [Cresap] criteria in a way that systematically discriminated against jobs performed mainly by women.’ ” Id. at 1380 (quoting report of Dr. Treiman).
Second, Dr. Treiman testified as to a study in which he analyzed each job classification using two criteria that he believed were objectively measurable: the amount of training and experience required, and whether the position entailed supervisory responsibility. Id. at 1381. After comparing jobs that require similar training and experience, and those that entail the same supervisory responsibility, Dr. Treiman again concluded that in 1967 the County “could not have applied the Cresap job-evaluation procedures in a good faith manner.” Id.
The County offered the testimony of two experts, Dr. David Jones, an industrial psychologist, and Dr. Joan Haworth, a labor economist and statistician. Dr. Jones testified about a study in which he tried to replicate the results of the 1967 Cresap process. Id. at 1384. This study, conducted in much the same way as Dr. Treiman’s first study, was intended to show that Dr. Treiman’s specific methodology was flawed. According to Dr. Jones, any disparity in salaries between male and female predominated jobs could be accounted for statistically by introducing a variable that represented the prevailing wages for those jobs in the regional job market. Id. Dr. Haworth testified that she had performed a similar study, and also concluded that any disparity in salaries was explained by market forces. Id. at 1390. Both experts therefore opined that the “salary grade differentials are not the product of sex discrimination but of an attempt to adjust salary grades so as to track market salaries and so as not to depart” from salaries that existed prior to the 1967 job classification process. Id. at 1391.
On August 24, 1992, the district court issued an opinion and order holding in part that AFSCME (i) failed to establish that the County intentionally discriminated against women when it set the pay scale for its employees in 1967, id. at 1392, 1412; and (ii) failed to show that the County intentionally discriminated in its conduct subsequent to the 1967 Cresap job classification process, id. at 1405, 1412. In arriving at these conclusions, the district court found that the studies on which AFSCME relied were deeply flawed. Dr. Treiman’s study attempting to reconstruct the Cresap process was found to be “a study in poor guidance, incomplete information, and inadequate time,” entitled to no weight. Id. at 1380. The court found that Dr. Michelson’s study, on which AFSCME relied to show that the County presently discriminated, had serious “methodological shortcomings,” id. at 1397, was flawed by “selective departures from [its] stated methodology,” id. at 1399, “deliberately omitted” relevant variables, id. at 1400, and failed to account for market forces,1 id. at 1401. On the other hand, the court found Dr. Haworth’s testimony (sponsored by the County), that differences in pay that developed after 1967 were the result of market forces, “reliable and credible,” and entitled to “great weight.” Id. at 1403.
After rejecting AFSCME’s more generalized claim that the County intentionally dis*649criminated in setting pay, the district court turned to AFSCME’s claims that the County intentionally set lower pay for three specific positions held predominantly by women. The district court found that AFSCME failed to show the requisite comparability between the jobs of a police communications operator and a fire communications technician, id. at 1408; or between the jobs of a clerical worker and a correction officer, id. at 1409-10. Thus, AFSCME could not establish a violation of Title VII or the Equal Pay Act with respect to these positions.
However, the court found that the work performed by police detention aides (predominantly women) was substantially similar to that performed by turnkeys (predominantly men); that police detention aides were paid less than turnkeys; and that “the salary disparity between these two groups is attributable to intentional discrimination.” Id. at 1409. The court noted that “this single instance of discrimination ... does not, without more, establish the broader claims of the plaintiffs in this action.” Id. On August 16, 1993, the district court ordered the County to pay the female police detention aides in the certified class a total of $1.6 million in back-pay for a fourteen year period.
No appeal was taken on any issue of liability-
After trial, the County moved for attorney’s fees and expert witness fees pursuant to 28 U.S.C. § 2000e-5(k). In a memorandum and order dated June 24, 1993, the district court found that the “plaintiffs’ pursuit of their claims was unreasonable and without foundation,” the standard for assessing fees against a losing plaintiff under Christiansburg, 434 U.S. at 422, 98 S.Ct. at 701, and therefore awarded the County $982,407.23 in attorney’s fees. AFSCME, 825 F.Supp. at 473-74. In concluding that AFSCME’s claims were frivolous, the court noted that the only evidence introduced to support AFSCME’s claim was expert testimony that was found to be flawed and unreliable, that AFSCME offered no other proof of discriminatory intent, and that AFSCME was a “major union” and “not economically or otherwise disadvantaged.” Id. at 471-73.
Expert witness fees were expressly made available to prevailing parties in a 1991 amendment to 42 U.S.C. § 2000e-5(k). See Pub.L. 102-166, Tit. I, § 113(b), 105 Stat. 1075,1079 (Nov. 21,1991). Because this case was tried prior to 1991, the availability of expert witness fees depended on whether the 1991 amendment is retroactive. At the time that the County’s motion for expert witness fees was before the district court (in 1993), the Supreme Court had granted certiorari in Landgraf v. USI Film Prods. to resolve the retroactivity issue. See 507 U.S. 908, 113 S.Ct. 1250, 122 L.Ed.2d 649 (1993) (granting certiorari). The district court deferred ruling on the County’s motion for expert witness fees pending the outcome of Landgraf. In 1994, the Supreme Court issued its decisions in Landgraf, 511 U.S. 244, 114 S.Ct. 1483, 128 L.Ed.2d 229 (1994) and a companion case, Rivers v. Roadway Express, Inc., 511 U.S. 298, 114 S.Ct. 1510, 128 L.Ed.2d 274 (1994). In an unpublished memorandum and order dated May 31, 1995, the district court described Rivers and Landgraf as holding that retroactive application of procedural rules is permissible, determined that the expert witness fee provision of Title VII is analogous to a procedural limitations provision because it applies to “secondary conduct — namely litigation conduct, not the alleged acts of discrimination,” and therefore held that the 1991 amendment applied retroactively to this case. Accordingly, the court awarded the County $550,974.66 in expert witness fees.
DISCUSSION
AFSCME appeals on two grounds. First, AFSCME contends that its claims were not “frivolous, unreasonable, or groundless” under Christiansburg, 434 U.S. at 422, 98 S.Ct. at 701, and that the district court therefore abused its discretion in awarding attorney’s and expert witness fees to the County. Second, AFSCME contends that, even if its claims were frivolous, an award of expert witness fees was improper because the 1991 amendment to 42 U.S.C. § 2000e-5(k) does not apply retroactively to this case. Because we conclude that AFSCME’s claims were not frivolous, and therefore no fees may be *650awarded under Christiansburg, we do not reach the retroactivity issue.
An award of fees under Title VII “is within the discretion of the trial court and will not be upset unless abused.” Carrion v. Yeshiva Univ., 535 F.2d 722, 727-28 (2d Cir.1976). As amended, section 706(k) of Title VII of the Civil Rights Act of 1964 provides:
In any action or proceeding under this subchapter the court, in its discretion, may allow the prevailing party, other than the Commission or the United States, a reasonable attorney’s fee (including expert fees) as part of the costs....
42 U.S.C. § 2000e-5(k). The clause “(including expert fees)” was added by the Civil Rights Act of 1991. See Pub.L. 102-166, Tit. I, § 113(b), 105 Stat. 1075, 1079 (Nov. 21, 1991).
Although the text of the statute does not distinguish between prevailing plaintiffs and prevailing defendants, the Supreme Court has held that a defendant is not entitled to an award of fees on the same basis as a prevailing plaintiff. See Christiansburg, 434 U.S. at 418-19, 98 S.Ct. at 699. The Court articulated “two strong equitable considerations” for permitting routinely an award of fees to prevailing plaintiffs that “are wholly absent” when a defendant prevails. Id. at 418, 98 S.Ct. at 699. First, “the plaintiff is the chosen instrument of Congress to vindicate ‘a policy that Congress considered of the highest priority.’ ” Id. (quoting Newman v. Piggie Park Enters., 390 U.S. 400, 402, 88 S.Ct. 964, 966, 19 L.Ed.2d 1263 (1968)). “Second, when a district court awards counsel fees to a prevailing plaintiff, it is awarding them against a violator of federal law.” Id. Mindful of those distinctions, the Court held that
a district court may in its discretion award attorney’s fees to a prevailing defendant in a Title VII case upon a finding that the plaintiffs action was frivolous, unreasonable, or without foundation, even though not brought in subjective bad faith.
Id. at 421, 98 S.Ct. at 700. Finally, the Court warned district courts to avoid “en-gag[ing] in post hoc reasoning by concluding that, because a plaintiff did not ultimately prevail, his action must have been unreasonable or without foundation.” Id. at 421-22, 98 S.Ct. at 700.
The Christiansburg standard for fee awards to prevailing defendants has been applied to civil rights cases arising under 42 U.S.C. § 1983, and to eases arising under other statutes that authorize fee awards. See Hensley v. Eckerhart, 461 U.S. 424, 429 n. 2, 103 S.Ct. 1933, 1937 n. 2, 76 L.Ed.2d 40 (1983) (action arising under § 1983); Hughes v. Rowe, 449 U.S. 5, 15, 101 S.Ct. 173, 178-79, 66 L.Ed.2d 163 (1980) (per curiam) (same); Rounseville, 13 F.3d at 632 (whether § 1983 defendant may be awarded fees “generally turns on whether the [plaintiffs] claim itself is clearly meritless”); Sassower v. Field, 973 F.2d 75, 79 (2d Cir.1992) (fees arguably warranted if “claim, though requiring jury consideration, was entirely insubstantial”), cert. denied, 507 U.S. 1043, 113 S.Ct. 1879, 123 L.Ed.2d 497 (1993).
We have indicated in particular cases what is meant by “frivolous, unreasonable, or without foundation.” In Gerena-Valentin v. Koch, 739 F.2d 755, 756-57 (2d Cir.1984), the plaintiff claimed that two municipal employees conspired to keep him off the ballot in an election for New York City councilman, in violation of the Voting Rights Act, 42 U.S.C. § 1973 et seq. Applying the Christiansburg standard, we approved a fee award to the prevailing defendants because the plaintiff had already litigated the issue (unsuccessfully) in state court, and because “at no time ... did [the plaintiff] attempt to produce any evidence whatsoever in support of his retaliation and conspiracy claim.” Id. at 761. In Eastway Constr. Corp. v. City of New York, 762 F.2d 243, 246 (2d Cir.1985), a general contractor charged the City of New York and others with violating the antitrust and civil rights laws by prohibiting it from contracting with other companies “engaged in City-financed reconstruction projects.” The effect of the City’s action was to put the plaintiff-contractor out of business. Id. Again applying Christiansburg, we approved a fee award to the defendant because the plaintiff “could not point to a deprivation of any single right conferred by federal law or the ... Constitution,” and had unsuccessfully challenged the City’s action in state court. Id. at 252. And in Faraci v. Hickey-Freeman Co., 607 F.2d *6511025, 1027-29 (2d Cir.1979), we approved a fee award against a Title VII plaintiff where the evidence of non-discrimination was “un-contradicted.” See also Harbulak v. County of Suffolk, 654 F.2d 194, 196-98 (2d Cir.1981) (approving fee award against Title VII plaintiff who alleged police officer violated his right to privacy by reaching into his car to serve summons); Prate v. Freedman, 583 F.2d 42, 47-48 (2d Cir.1978) (approving fee award against Title VII plaintiffs where “there was little or no chance that they would be permitted to attack [a prior consent decree and] judgment through a new suit”).
AFSCME asserted claims that were actionable under Title VII and had not previously been litigated, and offered proof as to each element of each claim. The County, of course, ultimately prevailed on most of AFSCME’s claims; but that alone is insufficient to justify an award of fees under the Christiansburg standard. After reviewing all the circumstances of this case and the proof introduced at trial, taken together, we are persuaded that AFSCME’s case was not “frivolous, unreasonable, or without foundation” so as to permit an award of fees under Title VII.
1. AFSCME Established A Prima Facie Case.
First, plaintiffs established a prima facie case of gender-based pay discrimination by introducing statistics tending to show an unexplained differential in wages. See Hazelwood School Dist. v. United States, 433 U.S. 299, 306-08, 97 S.Ct. 2736, 2741-42, 53 L.Ed.2d 768 (1977) (gross statistical disparities may prove prima facie case of discrimination); Teamsters, 431 U.S. at 339-40, 97 S.Ct. at 1856-57 (same). AFSCME’s two experts testified that, when all other relevant job characteristics were considered, there remained a substantial differential in the pay of men and women in the County. See AFSCME, 799 F.Supp. at 1379-80, 1396. One of the County’s two experts conceded that an unexplained pay differential existed:
There’s also still in all of these [job] specifications a gap between men[’s] and women’s salaries that has not been explained by the job specification variables.
See id. at 1401 (noting that Dr. Howarth confirmed “that there is indeed a gap between the salaries of men and women in Nassau County that is not explained by job specifications alone”).
In determining whether AFSCME’s claim was frivolous, the district court gave no weight to the prima facie ease, and described it as “meaningless.” 825 F.Supp. at 473. True, a prima facie case of discrimination may rest on a de minimis evidential showing. See O’Connor v. Consolidated Coin Caterers Corp., - U.S. -, -, 116 S.Ct. 1307, 1310, 134 L.Ed.2d 433 (1996) (a prima facie case could possibly be based on the “very thin evidence” of a 68 year-old being replaced by a 65 year-old). But here, AFSCME’s prima facie case was premised on several extensive statistical studies that undertook to show that the salary differential was not explained by the parameters of the Cresap process or by any other legitimate variables. Significantly, the existence of such an unexplained difference between salaries for men and women was conceded by the County. Under these circumstances, the pri-ma facie case at least created a factual question as to the cause of the gap in pay, and might have supported an inference of intentional discrimination. See Merrill v. Southern Methodist Univ., 806 F.2d 600, 607 (5th Cir.1986) (statistical evidence of wage disparity is probative of discrimination where “the employers unexplainedly departed from objective pay criteria they had adopted”). Indeed, had the County failed to respond to AFSCME’s prima facie case, the district court could have entered directed judgment against the County. See Teamsters, 431 U.S. at 361, 97 S.Ct. at 1867 (in Title VII class action, trial court may enter directed verdict against employer that fails to rebut prima facie case).
The district court therefore erred in affording AFSCME’s ample prima facie case no weight in the calculus under Christiansburg. See LeBeau v. Libbey-Owens-Ford Co., 799 F.2d 1152, 1160 (7th Cir.1986) (reversing grant of fees where EEOC established prima facie case and district court denied defendant’s motions for summary judgment and directed verdict), cert. denied, *652484 U.S. 815, 108 S.Ct. 67, 98 L.Ed.2d 31 (1987); EEOC v. Tarrant Distribs., Inc., 750 F.2d 1249, 1251 (5th Cir.1984) (per curiam) (affirming denial of fees where EEOC established prima facie case).
2. AFSCME Introduced Sufficient Evidence.
Because the district court had dismissed AFSCME’s disparate impact claim, AFSCME was required to establish that the County intentionally discriminated in setting pay. Establishing intent is difficult in any case, and even more so when the trial takes place more than twenty years after many of the events at issue. Faced with this challenge, AFSCME relied on statistical evidence introduced by its experts. See Teamsters, 431 U.S at 339 n. 20, 97 S.Ct. at 1856-57 n. 20 (“Statistics showing racial or ethnic imbalance are probative ... because such imbalance is often a telltale sign of purposeful discrimination.”). Those experts testified that the differential in salaries could not be explained fully by job characteristics, and therefore justified an inference that the disparity in pay between male- and female-predominated jobs was the result of intentional discrimination.
For many valid reasons, the district court — after a deft and sophisticated analysis of the complex data and methodologies, see AFSCME, 799 F.Supp. at 1378-1414-declined to credit that testimony, and ultimately concluded that AFSCME’s statistical studies were flawed. On appeal, the County contends that AFSCME’s decision to proceed to trial was unreasonable given the weak statistical evidence supporting its claim. But the court’s disregard of AFSCME’s statistical studies was based on determinations concerning technical methodology, integrity and validity that the district court arrived at with the benefit of effective cross-examination and the counterweight of the County’s experts and studies. AFSCME cannot be required to have anticipated these events and outcomes prior to trial. For example, the district court refused to credit the testimony of Dr. Treiman because (i) he was “extremely evasive and not entirely credible”; (ii) a subsequent report by him “recanted part of his [prior] conclusions”; and (in) “the court found his methodology in the conduct of [his study] to be so flawed as to render it unreliable and of no probative value in this case.” Id. at 1380. Under these circumstances, we cannot say — and the district court has not found — that AFSCME’s claim was frivolous at the outset. So the County may obtain fees only if “the plaintiff continued to litigate after it[s claims] clearly became” frivolous. Hughes v. Rowe, 449 U.S. at 15, 101 S.Ct. at 178-79 (quoting Christiansburg, 434 U.S. at 422, 98 S.Ct. at 701). The County points to no evidence that meets this test.
In determining whether a valid claim has been established, the assessment of evidence and testimony is of course within the province of the trier of fact. But a claim is not necessarily frivolous because a witness is disbelieved or an item of evidence is discounted, disproved or disregarded at trial. Compare EEOC v. Sears, Roebuck and Co., 114 F.R.D. 615, 631-32 (N.D.Ill.1987) (claim not frivolous where defects in statistical evidence not fully revealed until trial) with Carrion, 535 F.2d at 728-29 (pre-Christiansbwrg decision affirming fee award where plaintiffs testimony was “‘an unmitigated tissue of lies’ ” and whose lawsuit was “motivated by malice and vindictiveness” (quoting district court decision)). Generally, where evidence is introduced that, if credited, would suffice to support a judgment, fees are unjustified. See, e.g., EEOC v. Consolidated Serv. Sys., 989 F.2d 233, 238 (7th Cir.1993) (affirming denial of fees because, “had the judge believed the Commission’s witnesses, the outcome even of the disparate-treatment claim might have been different”) (Posner, J.); Johnson v. Allyn & Bacon, Inc., 731 F.2d 64, 74 (1st Cir.) (fees under Title VII inappropriate where result “could well have gone the other way if the district judge had believed” the plaintiff), cert. denied, 469 U.S. 1018, 105 S.Ct. 433, 83 L.Ed.2d 359 (1984); EEOC v. Kimbrough Inv. Co., 703 F.2d 98, 103 (5th Cir.1983) (affirming denial of fees where plaintiffs statistical evidence was flawed but where “some testimony and some analysis of hiring practices was valid probative evidence”).
*653In this case, AFSCME offered evidence as to each element of its claim. Cf. Gerena-Valentin, 739 F.2d at 761 (claim frivolous where no evidence of discrimination introduced); Faraci, 607 F.2d at 1027 (claim frivolous where evidence of non-discrimination was overwhelming). Thus, had the district court credited the evidence and testimony that AFSCME sponsored, AFSCME could have prevailed. The district court recognized that possibility when it stated:
It is easy to imagine that the plaintiffs could have trapped a victory in the web of statistical data spun by their expert witnesses were it not for a very able and experienced defense counsel who too mastered the statistics, regression analyses and complex computer data which enabled him to expose the deficiencies both in the experts’ testimony and in the studies upon which the plaintiffs so heavily relied.
AFSCME, 825 F.Supp. at 473. A case that could have resulted in victory for the plaintiffs but for “very able and experienced defense counsel” was not frivolous so as to justify a fee award, at least in the absence of evidence of intentional misleading by the plaintiffs.
3. AFSCME Obtained A $1.6 Million Judgment.
The award of fees here would arguably have presented a closer question had AFSCME recovered nothing on its claims. But AFSCME prevailed on its claim that the salary disparity between police detention aides and turnkeys “is attributable to intentional discrimination.” AFSCME, 799 F.Supp. at 1409. The district court awarded the plaintiffs a judgment in excess of $1.6 million on that claim. Although AFSCME failed on its broader claim — that the County’s entire pay scale was discriminatory — it is difficult to characterize as utterly meritless a lawsuit that resulted in a favorable judgment of more than $1.6 million.2 Certainly, a fee award might be warranted if one or more frivolous claims are accompanied by an unrelated claim that is meritorious. But the claim on which AFSCME prevailed was closely related to its other claims, and it was substantial by most measures. Under these circumstances, an award of fees was unwarranted under the Christiansburg standard, and the district court therefore abused its discretion in granting fees to the County.
CONCLUSION
The judgments of the district court awarding the County attorney’s fees and expert witness fees are reversed.
. Evidence was introduced that the County considered market forces in setting salary grades subsequentto 1967.
. The district court denied the County’s motion to dismiss AFSCME’s claims of unequal pay for equal work under the Equal Pay Act, and denied the County's motion to dismiss AFSCME’s disparate treatment claim under Title VII. See AFSCME, 609 F.Supp. at 707-11. The district court also denied the County's motion for directed verdict at the close of AFSCME’s case. These rulings are in some tension with the finding that AFSCME's claims were frivolous. See United States v. Mississippi, 921 F.2d 604, 609 (5th Cir.1991) (district court’s refusal to dismiss relevant to whether claim is frivolous); Sullivan v. School Bd. of Pinellas County, 773 F.2d 1182, 1189 (11th Cir.1985) (same).