The United States sued Fairfax County, Virginia, and certain governmental officials and agencies within the County1 for alleged violations of Title VII of the Civil Rights Act of 1964, 42 U.S.C. §§ 2000e et seq., the State and Local Fiscal Assistance Act of 1972, 31 U.S.C. § 1242 (Revenue Sharing Act), and the Omnibus Crime Control and Safe Streets Act of 1968, 42 U.S.C. § 3766 (Crime Control Act).2 It alleged and produced evidence at trial that the County *936pursued a pattern and practice of employment discrimination against blacks and women, in recruitment, hiring, assignments, and promotions. It also alleged and produced evidence at trial that the County used unvalidated testing devices with a disparate impact on blacks and women, that blacks and women were employed disproportionately in lower-paying and less desirable positions and that the County had refused to supply the Department of Justice with data necessary to a determination of whether the County was in compliance with the non-discrimination provisions of the Revenue Sharing Act, the Crime Control Act, and the regulations promulgated thereunder.3
After trial, the district court entered judgment largely for the County. Inexplicably, it failed to discuss and apparently failed to consider the government’s disparate impact case. With respect to the government’s disparate treatment case, it found that the County had discriminated against blacks in only two job categories and that it had discriminated against women in only one job category. Because of the County’s affirmative action program, the district court found it unnecessary to grant relief for the racial discrimination that it found, but it granted an injunction against the sex discrimination in the one job category. Finally, the district court declined to order the County to comply with record keeping regulations because it thought that the County would comply voluntarily. Both the government and the defendants appeal. In most part, we vacate the judgment and remand the case for further proceedings.
I.
As our introductory paragraphs indicate, this is both a disparate treatment and a disparate impact case. See Teamsters v. United States, 431 U.S. 324, 335, 97 S.Ct. 1843, 1854, 52 L.Ed.2d 396 (1977). This will necessitate separate discussion of the applicable law and separate discussion of some of the facts. Pertinent to both aspects of the case are the following facts:
A. General
The County has a work force of nearly 5,000 employees, divided into fifty departments which can be grouped into eight categories: (1) officials and administrators, (2) professionals, (3) technicians, (4) protective service workers, (5) para-professionals, (6) office and clerical workers, (7) skilled craft workers, and (8) service and maintenance workers. Within these categories, the County had 397 job classifications in which there were no or virtually no blacks or women; but, taking the categories as a whole, the County had in the year 1978 the following employees classified by race and sex:
Category Total White Black Male Female
1 Officials 149 142(95.3%) 4( 2.7%) 129(86.6%) 20(13.4%)
2 Prof. 782 735(94.0%) 25 ( 3.2%) 598(76.5%) 184(23.5%)
3 Tech. 536 498(92.9%) 24 ( 4.5%) 442(82.5%) 94(17.5%)
4 Prot. Serv. 1446 1350 (93.4%) 79 ( 5.5%) 1264 (87.4%) 182(12.6%)
5 Para-prof. 51 49 (96.1%) 2( 3.9%) 35 (68.6%) 16(31.4%)
6 Clerical 858 785 (91.5%) 45 ( 5.2%) 68 ( 7.9%) 790(92.1%)
7 Sk. Craft 575 526(91.5%) 30 ( 5.2%) 566(98.4%) 9( 1.6%)
8 Serv. & Maint. 599 427(71.3%) 167(27.9%) 577(96.3%) 22 ( 3.7%)
Totals 4996 4512(90.3%) 376 ( 7.5%) 3679(73.6%) 1317(26.4%)
B. Disparate Treatment Case
The United States produced two sets of statistical data to demonstrate that the percentages of blacks and women in the County’s work force were significantly dis-proportionate to the percentages of blacks *937and women in the available labor market. Thereby, the United States sought to prove a prima facie case of disparate treatment of blacks and women. Proof of a prima facie case would, of course, east the burden on the County to rebut the inference that it had- practiced purposeful racial and sexual discrimination.
First, the United States offered statistical data for the Washington, D. C. Standard Metropolitan Statistical Area (SMSA) based on the 1970 census and a 1974 Department of Labor Report. Significant disparities were proved. Defendants’ work force in 1978 was 7.5% black while the SMSA pool, measured by the 1970 census, was 24% black. As well, defendants’ work force was 26.4% female while the SMSA labor pool was 40.4% female.
Second, the government presented data showing the race and sex of persons who had applied for employment with the County (the “applicant flow data”). This evidence was followed by expert testimony that statistically significant disparities .existed in a wide range of job categories between the percentages of blacks and women in the 1978 applicant pool and the percentages of blacks and women hired by the County from 1974 through 1978.4 Specifically, the expert found disparities with respect to blacks in six of the eight job categories and with respect to women in five of the eight categories.
The United States attempted to prove disparate treatment of blacks and women through two other methods. The government showed that blacks and women were clustered in the lower-paying and less desirable job classifications. The service and maintenance category accounted for 44.41% of 1978 black incumbent employees, but only 9.45% of white incumbents. Of blacks hired from 1974 through 1978, 31.7% were assigned to the job classifications of public service workers or custodian in contrast to 4.9% of whites. Over 59% of defendants’ female employees in 1978 were employed in the office and clerical worker category, but only 1.85% of male employees were similarly assigned. Additionally, the government presented evidence of isolated instances of actual race and sex discrimination in the police department, the Department of Environmental Management, and the Sheriff’s Office.
The County’s defense to the government’s disparate treatment case consisted of its own statistical analysis and proof of its 1978 affirmative action plan. It prepared an analysis of the zip codes of 1978 applicants and determined the percentage of applicants from six geographical areas: Northern Virginia, the District of Columbia, suburban Maryland, the Middle Atlantic States, New York-Pennsylvania-Delaware, and the remainder of the United States. Figures from the 1970-census were used to determine the number of employed and experienced unemployed persons by race and sex from each of the six geographic regions, and based on these figures the County derived composite “availability” statistics for blacks and women in each of the job categories for these geographic areas. The evidence of actual hires in 1974-78 when compared to the composite available labor market showed significant disparities with respect to blacks for 1976 and 1977 in the protective service workers category and the service and maintenance workers category and with respect to females for 1974-78 in the service and maintenance workers category.
The County also presented evidence concerning its 1978 affirmative action plan. It introduced the plan and developed through testimony its efforts to increase the mailing list from 250 to 300 organizations, its “balanced certification” plan to certify minority applicants meeting minimum qualifications for interviews by the hiring agency, its policy of counting volunteer experience as part of an applicant’s qualifications, and its hiring goals. Because it had destroyed pre1978 applications, the County was unable to *938offer any evidence of the effectiveness of its 1978 affirmative action plan as compared to its pre-1978 recruitment efforts.
In deciding the government’s disparate treatment case, the district court rejected both the government’s SMSA data and its applicant flow data as measures of the County’s labor pool. The district court decided that the government’s SMSA statistics were not controlling because of the physical and economic distance between the County and the District of Columbia. The district court also found fatal flaws in the government’s applicant flow data, based upon the district court’s conclusion that the County’s affirmative action plan increased the percentages of women and black applicants in 1978. In addition, the district court declined to draw any inference of discrimination from the disproportionate number of females in the clerical category and the disproportionate number of blacks in the service and maintenance category. It concluded that “it is as reasonable to infer that these statistics are a result of preference as of discrimination.”
The district court chose to judge the County’s employment record against the County’s zip code analysis of its labor market. Even though that analysis revealed that the County had hired smaller percentages of blacks in two job categories5 than were present in the labor market, the district court concluded, based on the County’s affirmative action program, that equitable relief was unnecessary to correct racial discrimination. Because of a finding of discrimination against women in one job category,6 however, the district court did enjoin future sex discrimination.
C. Disparate Impact Case
In its disparate impact case, the United States presented evidence establishing that the County had used six employment tests, which were prerequisites for job classifications constituting approximately one-third of the County’s work force, that operated to exclude blacks from employment and promotion. Its evidence is summarized in the following table:
Coupled with these statistics, the government offered the testimony of an expert that the disparity in pass rates of blacks and whites was statistically significant.
The County offered no evidence that any of these tests had been fully validated to establish their relationship to the skills needed in the particular jobs. It did offer evidence that validation studies were under way for the entry level tests for firefighters and policemen. It sought through the testimony of its own employees to show that the firefighter II test and the test for clerical and clerical specialists adequately examined *939skills necessary for tasks performed on the job. Additionally, the County showed that the police corporal tests had been validated in another state. Finally, the County relied on its evidence in the disparate treatment aspect of the case, together with a government exhibit which was not offered in evidence,8 to prove that blacks were hired in proportions which indicated no disparate impact, contending that any disparate impact in the testing process was rectified by the lack of discrimination in the end result.
In entering its final decree, the district court did not discuss the disparate impact aspect of the case except to state that the validation studies for protective service entry-level tests were no more than “an excellent first step”, but that “the tests have not yet been validated.”
II.
A. The Disparate Treatment Case
The government undertook to prove its disparate treatment case largely through the use of statistical evidence showing a substantial racial and sexual imbalance in the County’s work force. Proof of such an imbalance “is often a telltale sign of purposeful discrimination; absent explanation, it is ordinarily to be expected that nondiscriminatory hiring practices will in time result in a work force more or less representative of the racial and ethnic composition of the population in the community from which employees are hired.” Teamsters v. United States, 431 U.S. at 340, n. 20, 97 S.Ct. at 1857. Of course, proof of a discriminatory motive is a necessary element of a disparate treatment ease, but statistics can establish a prima facie case, even without a showing of specific instances of overt discrimination. See, e. g., Hazelwood School District v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977); Barnett v. W. T. Grant Co., 518 F.2d 543, 549 (4 Cir. 1975).
The government’s SMSA statistical evidence showed wide disparities between the relatively small percentages of blacks and women in the County’s work force and the higher percentages of blacks and women in the County’s labor market. As well, the government’s applicant flow evidence demonstrated statistically significant disparities between the County’s labor force and its applicant pool for blacks in six job categories and for women in five job categories. By contrast, the County’s statistics showed significant disparities in the hiring of blacks in only two job categories and of women in only one job category. Thus, the degree of disparity between the percentages of blacks and women in the County’s work force and in its labor market hinges on the choice of a relevant labor pool. The critical question for decision, then, is: Which is the relevant labor pool, the government’s SMSA data, the government’s applicant flow data, or the County’s composite labor market derived from the study of zip codes?
It is to this question that we first direct our attention. While we think that the record supports the district court’s rejection of the SMSA statistics, we do not think that government’s applicant pool data should have been rejected. Concomitantly, we find the evidence of deficiencies in the County’s zip code analysis sufficiently compelling so that we do not think that the use of this measure of the applicable labor pool should be sustained.
The greatest disparities between the hirings of women and blacks and the County’s labor market appeared when the SMSA statistics were used as the relevant labor pool. The district court rejected the SMSA as a measure of the County’s labor market, and although we recognize that there was conflicting evidence, we do not consider the district court’s decision to be in error. As the government emphasizes, the County itself used the SMSA as the relevant labor *940market in three of its affirmative action plans and in a memorandum on minority police hiring. In addition, the County advertised job vacancies in metropolitan media, sent job announcements to organizations throughout the SMSA, and imposed no residence requirement for any job except county executive. But there was evidence that the distance between the County and the District of Columbia, where there is a substantial black population which would raise the percentage of blacks in the SMSA statistics, was so great that without adequate mass transportation, which does not exist, the two labor markets are separate and distinct. Another factor contributing to the separation and inhibiting a flow of applicants from the District of Columbia is the fact that wage rates in the County are lower than in the District. On balance, we perceive no error in the district court’s rejection of the SMSA statistics.
The district court also rejected the government’s second .method of establishing the County’s labor market, the applicant flow data. At the outset, we note that applicant data are normally highly relevant evidence of an employer’s labor market. See Hazelwood School District v. United States, 433 U.S. at 308 n. 13, 97 S.Ct. at 2741. Those who apply constitute the pool from which employees are selected.9
In this case, the applicant data from which the government prepared its statistics were available only for the year 1978 because the County had destroyed all of the applications for 1974-77. As a consequence, the government extrapolated the figures for 1974-77 from the 1978 applicant pool by assuming that the applicant pool for 1974-77 contained the same proportions of blacks and women as did the 1978 applicant pool. The district court found two related defects in the government’s extrapolated applicant flow data; we find, however, that neither of the supposed defects was serious enough to justify the district court’s decision to reject the applicant flow data, here the most salient proof of the County’s labor market.
First, the district court found the government’s figures suspect because the court thought that the County’s recent attempts at affirmative action probably resulted in larger percentages of women and black applicants in 1978 than in earlier years. We think that the district court’s assumption is not substantiated by the evidence. The 1977-78 affirmative action plan was not the County’s first; the County had had affirmative action programs since 1972. Some of the features of the County’s 1977-78 program were basically continuations of past programs, such as publicity and mailings to minority organizations. Moreover, some other affirmative action efforts had been discontinued by 1978, and some of the new features of the 1977-78 program, e. g., giving credit for volunteer work, would have little effect on recruitment. On balance, then, we see no basis for concluding that the 1978 applicant figures were significantly different from those of prior years.
We are likewise not persuaded by the other reason assigned by the district court for rejecting the government’s applicant flow data — that the 1978 data would penalize defendants because the County’s affirmative action programs would more likely attract a larger number of unqualified applicants than had applied in earlier years. We cannot assume that blacks and women attracted by an affirmative action plan are more likely to be unqualified than white males who apply. If such is the case, this possible defect in the statistics is a factual matter to be proved by defendants in rebuttal of the government’s prima facie case; it is not a factor which determines if a prima facie case has been proved. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802-O3, 93 S.Ct. 1817, 1824, 36 L.Ed.2d 668 (1973). Neither of these two supposed *941flaws, then, justify the district court’s decision to ignore the applicant flow data,
The district court’s decision to reject the government’s applicant flow data is the obverse of its decision that the County’s zip code analysis provided the most reliable measure of the County’s labor market. Unlike the district court, we perceive deficiencies in the County’s zip code statistics which distort them sufficiently, in our view, that they cannot supplant the applicant flow data as the appropriate measure of the County’s labor pool.
The County’s zip code analysis suffered from two fundamental defects. First, the zip code percentages which the County derived from its applications were applied to 1970 census data. That data is suspect and probably an inaccurate reflection of actual conditions in 1974-78. The 1970 census data recorded only the numbers of blacks and women in particular job categories during a period when the effects of racial and sexual discrimination in employment were undoubtedly more severe than in subsequent years. See Smith v. Union Oil Co., 17 FEP Cases 960, 967 (N.D.Cal.1977). Second, the County used only the population figures for the employed and the experienced unemployed blacks and women in the various job categories in 1970. Many of the County’s employment opportunities, e. g., entry-level police and fire, clerical, and maintenance jobs, surely require no prior experience, even if they do require an entrance examination. See Hazelwood School District v. United States, 433 U.S. at 308 n. 13, 97 S.Ct. at 2741; Teamsters v. United States, 431 U.S. at 337 & n. 17, 97 S.Ct. at 1855. Thus, the use of the County’s figures undoubtedly understated the available labor market for these jobs by including only blacks and women who held or who had held similar positions.10
In summary we think that, on the present record, the government’s applicant flow data was the appropriate measure of the available labor market for the period 1974-78. When resort is had to the applicant flow data, it is obvious that the government proved a more extensive prima facie case than the district court recognized. We do not undertake to define the exact contours of the case, however. That is more properly the function of the district court. And, the district court must also weigh what evidence the County may adduce in rebuttal to explain significant statistical disparities in the employment of blacks and women and to negative the inference of discriminatory intent.
While we express no view on the extent of the violation of Title VII which the district court may find on remand, we are constrained to comment on the limited relief the district court granted the victims it identified and its further refusal to grant a mandatory injunction to insure compliance with the record keeping requirements of law. In both respects, we think that the district court was in error.
To the extent that the district court finds racial discrimination, it is under a duty to render a decree which will both eliminate past discrimination and bar discrimination in the future. Albemarle Paper Co. v. Moody, 422 U.S. 405, 418, 95 S.Ct. 2362, 2372, 45 L.Ed.2d 280 (1975). It is commendable that the County is continuing its affirmative action programs, although there was some evidence that the goals may soon be reduced. But, in any event, as we said in Barnett v. W. T. Grant Co., 518 F.2d at 550, “a court cannot abdicate to defend*942ants’ good faith its duty of insuring removal of all vestiges of discrimination.”
Thus, we think that the district court should have granted injunctive relief against future discrimination. In granting injunctive relief, it should both have required compliance with the record keeping and disclosure requirements of existing law. See EEOC v. Rogers Brothers, Inc., 470 F.2d 965 (5 Cir. 1972),11 and imposed requirements for periodic reports to enable it to monitor compliance with its decree. Finally, if proof is offered of identifiable economic injury to blacks or women, or both, who have suffered from the County’s discriminatory practices, it should grant back pay or retroactive seniority or both. Teamsters v. United States, 431 U.S. at 361-62, 97 S.Ct. at 1867-68; Albemarle Paper Co. v. Moody, 422 U.S. at 421, 95 S.Ct. at 2373; Hill v. Western Electric Co., 596 F.2d 99, 104 (4 Cir. 1979), cert. denied, 444 U.S. 929, 100 S.Ct. 271, 62 L.Ed.2d 186 (1979); Sledge v. J. P. Stevens & Co., 585 F.2d 625, 637 (4 Cir. 1978); Robinson v. Lorillard Corp., 444 F.2d 791, 803-04 (4 Cir. 1971). At this time and on this record, we do not pass on the government’s contention that the district court impose hiring goals and timetables. We only call attention to Sledge v. J. P. Stevens & Co., supra, where we have said that hiring quotas should be imposed only in the most extraordinary circumstances and where there is a compelling need.
B. The Disparate Impact Case
In order to prevail in a disparate impact case, a plaintiff need not show that a defendant acted with a discriminatory purpose. He need only show that a facially neutral employment practice has a discriminatory impact on members of a protected group. If that showing is made, use of the procedure is unlawful unless it is shown to be valid or otherwise required by business need. See, e. g., Dothard v. Rawlinson, 433 U.S. 321, 329, 97 S.Ct. 2720, 2726, 53 L.Ed.2d 786 (1977); Griggs v. Duke Power Co., 401 U.S. 424, 430-32, 91 S.Ct. 849, 853-854, 28 L.Ed.2d 158 (1971).
In the instant case we think that the government proved a prima facie case of disparate treatment of blacks in the County’s use of hiring and promotion tests having a severe adverse impact on blacks in job categories which included 30% of the work force. Because defendants failed to establish, with possibly one exception, that the tests were properly validated to measure job performance, the use of these tests would appear to be in violation of Title VII. Washington v. Davis, 426 U.S. 229, 247 n. 13, 96 S.Ct. 2040, 2051, 48 L.Ed.2d 597 (1976); Albemarle Paper Co. v. Moody, 422 U.S. 405, 95 S.Ct. 2362, 45 L.Ed.2d 280 (1975); Griggs v. Duke Power Co., 401 U.S. 424, 91 S.Ct. 849, 28 L.Ed.2d 158 (1971).
We decline, however, the Government’s invitation to reach a final decision on this aspect of the case. There are many unresolved factual aspects of the disparate impact case, most particularly with regard to the proof offered by the County to rebut the government’s showing of adverse impact. The district court has made no factual findings. We recognize that the County, except possibly with regard to the test for police corporal, offered no evidence of professional validation of the tests it employs.12 *943Usually the starting point in proof of validity is evidence of a thorough job analysis. See Vulcan Society v. Civil Service Commission, 360 F.Supp. 1265, 1274 (S.D.N.Y.), aff’d, 490 F.2d 387 (2 Cir. 1973). And this proof is followed by demonstration that the test accurately and fairly measures the knowledge, skills, and abilities needed for successful performance of the job. See United States v. City of Chicago, 573 F.2d 416 (7 Cir. 1978); Kirkland v. New York State Department of Correctional Services, 374 F.Supp. 1361 (S.D.N.Y.1973), aff’d in relevant part, 520 F.2d 420 (2 Cir. 1975), cert. denied, 429 U.S. 823, 97 S.Ct. 73, 50 L.Ed.2d 84 (1976); Vulcan Society v. Civil Service Commission, supra. Nonetheless, with these legal criteria as a guide, we think that the initial exploration of the factual mazes of the disparate impact case should be undertaken by the district court and not by us.13
Based upon the foregoing, we vacate the judgment of the district court and remand the case for further proceedings consistent with what we have stated.
VACATED AND REMANDED.
. Collectively, defendants will be sometimes referred to as the “County.”
. Because Title VII standards also govern discrimination claims raised under the Revenue Sharing Act and the Crime Control Act, this opinion will discuss only Title VII. See United States v. New York, 82 F.R.D. 2, 4 5 (N.D.N.Y. 1978); United States v. Baltimore County, 17 EPD ‘ 8509 (D.Md.1978).
. 31 C.F.R. g 51.59; § 4, Department of Justice Employee Selection Guidelines; 28 C.F.R. § 42.-106.
. In violation of the record keeping regulations of the Revenue Sharing Act and the Crime Control Act, defendants had destroyed pre1978 applications for employment. As a consequence, the government’s expert, assuming that the rate remained constant, used the 1978 rate of applications from blacks and women and applied it to previous years.
. The zip code analysis uncovered discrimination against blacks during 1976 and 1977 in the protective services and the service and maintenance categories.
. Sex discrimination was found in the service and maintenance category for the years 1974 78.
. This exhibit contained statistics purporting to represent the number of black applicants. The statistics were furnished to the government by the county’s personnel office. The government decided not to introduce the exhibit into evidence, allegedly because the statistics were unreliable.
. Of course, if the victims of discrimination have been discouraged from applying, the applicant pool data may not adequately measure the percentages of women and blacks actually present in the labor market. See Dothard v. Rawlinson, 433 U.S. 321, 330, 97 S.Ct. 2720, 2727, 53 L.Ed.2d 786 (1977). No such challenge has been made to the applicant pool data in this case.
. In addition, the County’s study employed zip codes which included areas of distant Fauquier and Rappahannock Counties.
It is significant also that, because of illness in his family, the preparer of the zip code analysis did not testify. His supervisor, who sponsored the exhibits, acknowledged numerous weaknesses in the data. The availability figures which were developed were given in terms of the categories of employment stated in the text, but they were derived from census figures which do not employ those categories. The supervisor could not explain or justify how the data was translated. Indeed, she could not identify the source for any data included in the exhibit.
. Normally, a district court has discretion to deny injunctive relief to require record keeping, when the court is convinced that the transgressor will keep adequate records in the future. In this case, however, we find that the district court exceeded its discretion by refusing to issue a record keeping injunction. The County had a long history of inadequately maintaining employment records; it had stored applicant records in a shoe box and routinely destroyed them. This lax attitude toward compliance with the record keeping regulations made exact proof in this case more difficult, because the applicant data for 1974-77 were not available. The County began adequate record keeping practices only after the instigation of this litigation. On these facts, the district court should have taken steps to insure that the County would fulfill its record keeping obligations.
. The defendants produced a validation study conducted in North Carolina for the police corporal test.
. Consequently, we do not address the County’s “bottom line” defense, by which it contends that sufficient numbers of blacks were hired even if the employment tests, which allegedly constitute only a part of the hiring process, were discriminatory. We pass no judgment on the theoretical validity of this defense and leave it in the first instance to the district court. The district court should examine the factual aspects of the County’s defense closely, in light of the government’s insistence that the figures on which it is based are suspect. See note 8 supra.
. The figures for the 1974 1975 police officer examination are for Fairfax County, Arlington County, and the City of Alexandria. The same examination was given for each jurisdiction.