The Equal Employment Opportunity Commission (EEOC) instituted this action on a complaint charging that defendant American National Bank (ANB or Bank) had engaged in a pattern or practice of *1181racially discriminatory hiring practices from 1969 to 1975, in violation of Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq. The district court found that the static work force statistics submitted by the EEOC revealed a continuous underrepre-sentation of blacks in defendant’s work force during the relevant period, both generally and in the specific job categories of officials and managers and of office and clerical personnel, and therefore constituted prima facie proof of a pattern or practice of discrimination. It then held that prima fa-cie case rebutted, however, by ANB’s applicant flow data and a standard deviation analysis of the static work force statistics. The court then examined each of ANB’s hiring practices that the EEOC had alleged to be discriminatory, and found them to be legitimate business practices and nondiscriminatory in effect, both separately and in combination. Additionally, the court evaluated 31 claimed examples of discrimination in specific hiring decisions, and found that none showed a denial of employment because of race. Concluding that the EEOC had therefore failed to prove a pattern or practice of discrimination, the court dismissed the suit. It then ruled that the EEOC had brought a frivolous claim and pursued it in bad faith, justifying an award of costs and attorneys’ fees to defendant. In a later order, the court determined the amount of reasonable attorneys fees to be that submitted by ANB, in the amount of $106,084.75.
We agree with the district court’s ruling that the EEOC’s statistical proof made out a prima facie case of discrimination, but, with the exception of one employment category, we conclude that the court erred in holding that defendant’s rebuttal evidence was sufficient to overcome the prima facie case. Because the proof of a pattern or practice of discrimination thus stands unre-butted as to all but the one category on a proper legal analysis of the total evidence, we reverse and remand for the determination of appropriate relief.
I
This case began with a charge of discrimination filed with the EEOC in 1969 by a rejected black applicant, Sandra Holland. Ms. Holland alleged that she had been refused employment at ANB’s Suffolk branch. The EEOC conducted an investigation in 1970, and found that Ms. Holland’s application at the Suffolk branch could not be located, though it did find an earlier application at the Portsmouth branch. The EEOC issued its formal “determination” in March 1974, finding reasonable cause to believe that ANB had engaged in discriminatory hiring practices. The determination discussed hiring figures only for ANB’s Suffolk branch, and in a footnote rejected information about the hiring of blacks in ANB’s Portsmouth branches as “irrelevant to employment at its Suffolk branch.” App. 1523. Conciliation efforts were unsuccessful, and Ms. Holland was issued a right-to-sue notice in August 1974, but declined to file suit. The EEOC then instituted this action in January of 1976, charging a pattern or practice of discrimination in both the Suffolk and Portsmouth branches.
In August of 1976, the district court granted defendant’s motion for summary judgment and dismissed the action for laches. This court vacated and remanded the case, in EEOC v. American National Bank, 574 F.2d 1173 (1978). Applying Occidental Life Insurance Co. v. EEOC, 432 U.S. 355, 97 S.Ct. 2447, 53 L.Ed.2d 402 (1977), we held the action not barred because no federal statute of limitations applied to suits by the EEOC once jurisdiction over the complaint is properly obtained, and noted that Congress did not intend state statutes of limitation to apply to such suits. Additionally, we held that the EEOC was entitled to bring suit on the broader pattern or practice allegation, founded on its reasonable cause determination and conciliation efforts. Any prejudice to the Bank resulting from the loss of evidence relating to Sandra Holland’s charge did not affect its ability to defend the pattern or practice suit. 574 F.2d at 1175-76.
The case was tried over a four-day period in December 1978. The EEOC presented *1182statistical evidence designed to show gross disparities between ANB’s work force and the relevant labor pool of qualified blacks during the period charged. Further evidence was offered to prove ANB’s use of subjective selection criteria and other discriminatory hiring practices. Finally, the EEOC presented 52 specific instances of black applicants who were allegedly not hired because of their race. (Thirty-one of these testified at trial).
The EEOC’s statistical evidence, presented in the form of stipulated exhibits, compared on a static basis the racial composition of ANB’s work force in its Suffolk and Portsmouth branches from 1968-75 to that of the general available work force in the relevant market areas.1 It then compared the racial composition of specific job categories at both branches — officials and managers, office and clerical workers, and service workers — with the proportion of qualified blacks in the population, for each category.2 The statistics revealed, as the district court found, that in the Suffolk branches there were no blacks employed as officers or managers during the eight-year period, though blacks comprised 8-10.9% of the qualified labor market. In its twenty clerical positions, ANB-Suffolk employed one black in the years 1970 and 1973, and two in 1971 and 1972. There were no blacks employed in this category during the other relevant years, as compared to a qualified labor market that was 10.3-22.5% black. In the seven ANB-Portsmouth branches, there were no black managers from 1968-1972, and one for each of the remaining three years. The qualified work force during this period was 4.8-6.9% black. In the office and clerical categories at Portsmouth, the Bank’s work force was from 0 to 6.5% black from 1968-1974, and 9.3% in 1975, compared with a qualified work force that was 13.9-21.5% black. Only in the service worker category has the percentage of black bank employees been greater than the percentage of blacks in the available service worker force' in the relevant labor market area. Durihg the entire period, the service work force in both branches was at least 75%, and often 100% black.
The evidence as to the Bank s hiring procedures 3 was also stipulated by the parties. Applicants had to appear and submit an application form which would be retained on file and considered for vacancies for up to six months. The application form sought information on the individual’s education, work experience, skills, references, desired position and salary, and relatives or friends at the Bank. Before 1973, applicants were required to take the Wonderlic Personnel *1183test; until May 1975, a high school diploma was required for a clerical position. Application forms would be submitted to a designated employee; usually a different official would screen the applications and conduct interviews when a vacancy occurred. Until 1975, when the bank began maintaining records for determining compliance with affirmative action plans, no notation was made of the race of an applicant. There may have been clues, however, in their addresses and educational history.
Bank officials testified that when a vacancy developed, they would review the most recent applications first, and usually would not reach applications submitted more than 30 days previously. In a few cases, when they did not find a qualified applicant on file, they would advertise the vacancy. After reviewing applications, officials would select a few applicants to interview. There were no written job descriptions or criteria for hiring. The interviewers, all of whom were white, would evaluate the applicant’s ability to communicate, maturity, personality, and physical appearance. The interviewer’s decision to accept or reject an applicant was generally controlling. When ANB became particularly interested in hiring an applicant, prior employers were contacted, and character references were sometimes checked. Prior to 1975, ANB occasionally made retail credit checks on final applicants. Nearly half of the hires during this period had listed friends or relatives among bank employees.
As part of its nonstatistical evidence, the EEOC presented the testimony and written statements of 31 black applicants who had been rejected for positions at ANB. The EEOC stipulated that its case as to individual examples of discrimination would be limited to those 31 applicants who were able to testify. Twenty-one additional applicants were identified by the EEOC as having been discriminated against. Although these particular cases were not submitted to the court on the pattern or practice of discrimination issue, the court considered them in deciding the attorney’s fees issue.
The Bank, in rebuttal, submitted applicant flow data purporting to compare the number of blacks and whites hired with those who submitted an application during 1969-1975.4 These figures revealed that in the Suffolk branches during this time, blacks submitted 25.1% of the applications and represented 17.9% of those hired, while whites submitted 74.9% of the applications and constituted 82.1% of the hires. For the Portsmouth branches, only 1975 data was available. They indicated that blacks constituted 23.3% of the applicants and 11.1% of the hires, while whites were 76.1% of the applicants and 88.9% of the hires. These statistics did not separate the applicants by job category, but combined the service worker hires with the clerical and managerial hires.
Before ruling on the substantive issues under Title VII, the court considered the preliminary issue whether it had jurisdiction to hear the claims. Although finding that the initial charge by Ms. Holland was valid and the EEOC’s broader investigation of the general policies and practices of ANB’s Suffolk branches relevant to the initial charge, the court concluded that the activities of the Portsmouth branches were not the subject of the EEOC investigation nor were they included in the reasonable cause determination of March 11, 1974. Consequently, the court ruled that it lacked jurisdiction to hear that part of the case concerning the allegedly discriminatory practices and the individual claims of discriminatory hiring at the Portsmouth branches. Anticipating that the case would be appealed, the court nevertheless included the Portsmouth branches in its analysis of the merits.
Addressing the merits, the court first considered the static work force statistics presented by the EEOC, comparing the percentage of blacks in defendant’s work force with the percentage of blacks available in the population, using both general work force and special qualifications work force figures. From each set of figures, the *1184court found that blacks were underrepresented in ANB’s work force. Under Hazelwood School District v. United States, 433 U.S. 299, 97 S.Ct. 2736, 53 L.Ed.2d 768 (1977), the court therefore found that the EEOC had presented prima facie proof by statistical means of a pattern or practice of discrimination at ANB’s Suffolk and Portsmouth branches.
The court then evaluated ANB’s rebuttal evidence consisting of its applicant flow data for the charged period, and the EEOC’s nonstatistical evidence related to ANB’s screening and hiring practices, and the 31 cases of unsuccessful black applicants who testified at trial. The applicant-flow data and a standard deviation analysis of the static work force statistics were found sufficient in combination to rebut the EEOC’s prima facie case. Alternatively, though without discussion, the court found that too few hiring decisions were made to justify an interference of discrimination. After an extensive analysis of each of the hiring practices shown by the EEOC, the court concluded that neither separately nor together were these practices discriminatory but rather served legitimate business purposes. Finally, the court found that none of the 31 unsuccessful black applicants were victims of racial discrimination in hiring. On balance, then, the court concluded that the EEOC’s statistical evidence of a prima facie case was substantially rebutted. Conceding that ANB’s work force was virtually all white prior to 1969 and racially imbalanced from 1969-1975, the court concluded that the imbalance was a product of pre-1969 activities and therefore not viola-tive of Title VII. For reasons that follow, we find error requiring reversal in the court’s legal and factual analysis of the evidence.
II
Initially we consider and reject the district court’s conclusion that it had no jurisdiction over that part of the suit concerning ANB’s Portsmouth branches.
The 1972 amendments to Title VII gave the EEOC broad enforcement powers. Section 706 enabled the EEOC, after a charge alleging discrimination had been filed with the EEOC and conciliation efforts with the respondent had failed, to bring suit in its own name. 42 U.S.C. § 2000e-5(f)(l). Although Congress declined to expand the EEOC’s powers to the extent exercised by the NLRB, broad litigation powers were added to preexisting powers to enable the EEOC “to vindicate the public interest.” EEOC v. Kimberly-Clark Corp., 511 F.2d 1352, 1361 (6th Cir. 1975). Consistent with this expansion of the EEOC’s role to include enforcement as well as conciliation, EEOC v. Cleveland Mills Co., 502 F.2d 153, 155-56 (4th Cir. 1974); EEOC v. Kimberly-Clark Corp., 511 F.2d at 1357, the 1972 amendments were intended to enable the EEOC to correct “public or ‘societal’ wrongs, [to be] the public avenger by civil suit of any discrimination uncovered in a valid investigation and subjected to conciliation under the Act.” EEOC v. General Electric Co., 532 F.2d 359, 373 (4th Cir. 1976). We determined in General Electric that the EEOC had standing, after 1972, to bring suit in its own name on “discrimination stated in the charge itself or developed in the course of a reasonable investigation of that charge, provided such discrimination was included in the reasonable cause determination of the EEOC and was followed by compliance with the conciliation procedures fixed in the Act.” Id. at 366 (italics omitted). In giving the EEOC the right to sue in its own name, Congress signalled its desire “[to eliminate] employment discrimination as a national evil” rather than provide solely for the redress of private interests. To confine standing of the EEOC to that of an individual complaining party would nullify the legislative intent of the 1972 amendments. Id. at 373.
The EEOC’s new role as an enforcer, however, was not intended to diminish its role as conciliator. Title VII establishes a comprehensive administrative procedure whereby the parties involved, including the EEOC, have the fullest opportunity to resolve charges of discrimination without resorting to the courts. See 42 U.S.C. *1185§ 2000e-5(b), (f)(1). Before the EEOC may bring a civil action under Title VII against an employer there must be a charge filed with the EEOC, notice of the charge to the employer, investigation by the EEOC, a determination of reasonable cause, and an effort at conciliation. EEOC v. Raymond Metal Products Co., 385 F.Supp. 907, 916 (D.Md.1974), aff’d in relevant part, 530 F.2d 590 (4th Cir. 1976). Two important purposes are served by this preadjudicative administrative procedure: first, the employer is fully notified of the violation alleged by the charging party; and second, the EEOC has the opportunity to consider all the charges and to attempt their resolution through conciliation and voluntary compliance. Id.
The 1972 amendments, while enlarging the powers of the EEOC to include enforcement, retained the previous emphasis on administrative resolution and conciliation of charges. Interpretations of the 1972 amendments emphasize the sequential nature of the process. “[E]ach step in the Commission’s administrative process is designed to be a prerequisite to the following step and, ultimately, to suit.” EEOC v. E. I. DuPont de Nemours & Co., 373 F.Supp. 1321, 1336 (D.Del.1974), aff’d, 516 F.2d 1297 (3d Cir. 1975). Conciliation, the final step in the process, can therefore be regarded as a condition precedent to the EEOC’s power to sue. See EEOC v. Allegheny Airlines, 436 F.Supp. 1300 (W.D.Pa.1977), and legislative history of the 1972 amendments quoted therein.
The law on the question of the permissible scope of a complaint filed by the EEOC pursuant to the 1972 amendments has focused on the role of the EEOC as investigator and the statutory preference for administrative resolution of charges of discrimination. Proper investigation and conciliation of charges are characterized as jurisdictional prerequisites to a suit by the EEOC on a particular claim. But see EEOC v. Westvaco Corp., 372 F.Supp. 985, 991 (D.Md.1974) (EEOC’s failure to make a timely reasonable cause determination and to attempt conciliation before filing suit goes to whether a claim has been stated rather than to subject matter jurisdiction.)
The district court in this case dismissed the EEOC action insofar as it related to the Portsmouth branches of ANB and the individuals allegedly discriminated against at those branches. Characterizing the Portsmouth charges as “new discrimination” not included in the original charge by Sandra Holland, the court analyzed whether or not these charges could be included in the EEOC civil complaint under the test set forth in General Electric. Finding that the EEOC investigation and reasonable cause determination concerned only the allegations of discrimination at the Suffolk branches, the court concluded it was without jurisdiction to hear the Portsmouth charges.
We disagree with the district court’s analysis and conclusion on this issue. The question was not whether the court had jurisdiction over “new” charges of discrimination brought for the first time by the EEOC in its civil complaint. The crucial issue was instead whether the district court had jurisdiction over the same charges of discrimination against a single defendant, expanded to include the same practices at all its branch offices when the original charge and investigation focused on one city but where there was common ownership and control over branches in both that city and a nearby city, and where the challenged hiring practices for all branches were similar. We conclude that jurisdiction over charges pertaining to all branches of ANB was proper in this case because there was, through the EEOC’s investigation and attempted conciliation with regard to Suffolk, adequate notice to the defendant of the practices under investigation and ample opportunity for conciliation concerning those practices. Had the conciliation effort been successful, given the common control over and similar practices at the two cities’ branches, whatever changes were to be instituted at the Suffolk branches would no doubt logically and necessarily have been made at the Portsmouth branches as well.
Our resolution of this issue does not disturb the rule announced in General Electric *1186and since adhered to by this court. See EEOC v. Chesapeake & Ohio Ry., 577 F.2d 229, 231-32 (4th Cir. 1978). As we noted above, completion of the full administrative process is a prerequisite to the EEOC’s power to bring suit in its own name. This power of enforcement is bottomed on an administrative scheme giving every employer charged with discrimination under Title VII the opportunity to correct its fault out of court. See Macon v. Bailar, 451 F.Supp. 140, 142 (E.D.Va.1978), citing Alexander v. Gardner-Denver Co., 415 U.S. 36, 44, 94 S.Ct. 1011, 1017, 39 L.Ed.2d 147 (1974). Consistent with this scheme, we require that a particular charge of discrimination be the subject of the reasonable cause determination and conciliation before being subject to suit by the EEOC. This requirement, for example, protects an employer charged in the reasonable cause determination with race discrimination in hiring against being surprised by a subsequent suit including charges of race discrimination in layoffs or promotion, or sex discrimination. There would have been no prior notice to the employer that practices relating to these charges were suspect nor an opportunity for the employer to remedy the problems out of court.
This case, however, does not present such a situation. There is but a single charge of race discrimination in hiring. ANB is a single employer with operations at multiple locations, all subject to unified supervision and control and using similar hiring practices. 21 F.E.P. Cases at 1541. But cf. Stastny v. Southern Bell Telephone & Telegraph Co., 628 F.2d 267 (4th Cir. 1980) (importance of local autonomy and discrete labor pools in adjudicating liability); Hill v. Western Electric Co., 596 F.2d 99 (4th Cir. 1979) (same). Although the notice of the original charge by Sandra Holland was sent to Donald King, Assistant Vice President of ANB Suffolk, the receipt of the notice was acknowledged by Joseph M. Fail, Vice President at the main office in Portsmouth. App. 1518, 1519. Also, the EEOC’s notice of failure of the conciliation effort was sent to ANB’s attorney, id. at 1526, who presumably represented not only the Suffolk branches but the entire organization. ANB’s officers must therefore be charged with full awareness of the charges and the issues to be resolved in conciliation and thus fully aware of the practices challenged by the EEOC.5 There was therefore no possibility for prejudicial surprise of the sort that has properly been held to preclude proof of later added charges in other situations.
We conclude, therefore, that the district court did have jurisdiction over the Portsmouth claims and should have considered proof related to them.
Ill
We turn now to the district court’s analysis on the merits. As indicated in our gen*1187eral summary of the course of proceedings, the district court specifically concluded that, based upon the static work force statistical evidence, there was prima facie proof of a pattern or practice of race discrimination in hiring:
[T]he fact [is] that from 1969 to 1975, blacks were underrepresented in defendant’s work force, not only in the work force generally but also in the specific categories of officials and managers and office and clerical personnel. The Court cannot agree with defendant that these comparisons are “irrelevant.” Hazelwood School District v. United States, 433 U.S. 299, [97 S.Ct. 2736, 53 L.Ed.2d 768] 15 F.E.P. Cases 1 (1977), held that comparative statistics showing gross statistical disparities are prima facie proof of a pattern or practice of discrimination. Id. at 308-09 [97 S.Ct. at 2741-42], The Court finds, therefore, that the statistical evidence presented here by the EEOC is prima facie statistical proof of a pattern or practice of discrimination.
21 F.E.P. Cases at 1552 (footnote omitted). The court then concluded, however, after analyzing the defendant’s applicant flow data and applying a standard deviation analysis to both the static work force statistics and the applicant flow data, that the inference of a pattern or practice of race discrimination had been neutralized. Id. at 1560. Having thus concluded that the statistically based prima facie case had been effectively rebutted, the court then turned to independent consideration of the EEOC’s nonstatistical evidence offered to show a discriminatory pattern or practice. Concluding that this evidence would not support an inference of discrimination, the court was led inexorably to the ultimate conclusion that the EEOC had failed to carry its burden of proof. 21 F.E.P. Cases at 1584.
For reasons that follow, we conclude that in a number of critical respects the district court’s analysis of the evidence was flawed by specific failures correctly to apprehend or to apply controlling legal principles developed by the Supreme Court for analyzing the evidence in this type case. These misapprehensions and misapplications constituted errors of law which invalidate with one exception the court’s ultimate conclusion that the EEOC failed to prove the pattern or practice of discrimination as charged and properly under consideration. We further conclude that under a legally correct analysis of the essentially undisputed historical facts in evidence, a prima facie case was established and except with respect to managerial employees in the Suffolk branches not rebutted or avoided, and that in consequence a pattern or practice of racial discrimination in hiring was proved.
Specifically we find errors of law in the weight apparently assigned by the court to a standard deviation analysis of certain of the statistical evidence; in the significance which the court assigned to ANB’s applicant flow statistics as rebutting evidence; and in the way in which the court treated the relationship between the statistical and nonstatistical evidence offered to prove the discriminatory pattern or practice charged. To show why, we briefly summarize the controlling principles for analysis of proof in this case, and then indicate wherein we think the district court so far misapprehended or misapplied these as to make erroneous its ultimate conclusion that the EEOC had failed to prove a pattern or practice of discrimination violative of Title VII.
The controlling principles are those embodied in the proof scheme developed by the Supreme Court for assessing claims of patterns or practices of disparate treatment, principally in International Brotherhood of Teamsters v. United States, 431 U.S. 324, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977) and Hazel-wood. While designed merely to provide a “sensible, orderly way to evaluate the evidence,” Furnco Construction Corp. v. Waters, 438 U.S. 567, 577, 98 S.Ct. 2943, 2949, 57 L.Ed.2d 957 (1978), and not intended as an inflexible, rigid framework for all cases, id. at 575, 577, 98 S.Ct. at 2948-49, the basic design of this proof scheme is perfectly and appropriately adaptable to the evidence presented in the instant case.
*1188In the type case we consider, the ultimate issue is whether an employer regularly and purposefully treats or has treated blacks less favorably than whites and whether this disparate treatment is racially motivated. Teamsters, 431 U.S. at 335, 97 5. Ct. at 1854. The EEOC of course bears the initial burden of making out a prima facie case of discrimination. This prima facie showing may in a proper-case be made out by statistics alone, Teamsters, 431 U.S. at 339, 97 S.Ct. at 1856; Hazelwood, 433 U.S. at 307-08, 97 S.Ct. at 2741-42; Barnett v. W. T. Grant Co., 518 F.2d 543, 549 (4th Cir. 1975), or by a cumulation of evidence, including statistics, patterns, practices, general policies, or specific instances of discrimination. See Barnett v. W. T. Grant Co., 518 F.2d at 549; Brown v. Gas-ton County Dyeing Machine Co., 457 F.2d 1377, 1382 (4th Cir. 1972). If the prima facie case is established by the EEOC’s statistical evidence, the employer may nevertheless rebut it, dispelling the inference of a general policy of discrimination, by “demonstrating that the Government’s proof is either inaccurate or insignificant.” Teamsters, 431 U.S. at 360, 97 S.Ct. at 1867. One way that static work force statistics revealing gross disparities can be shown nevertheless to be “insignificant” is by showing that the disparities are mainly attributable to pre-Act rather than to post-Act employment actions. This may be done in two basic ways: by focusing on the static work force statistics and purging them of all pre-Act employment actions so that only post-Act actions remain for assessment; or, more commonly, by focusing on post-Act employment decisions, and showing either that considered alone, they affirmatively reveal nondiscrimination in the post-Act period or that they are insufficient in number to support an ultimate conclusion of a post-Act discriminatory pattern or practice.6 Hazelwood, 433 U.S. at 309, 313, 97 S.Ct. at 2742, 2744.
In the instant case, the EEOC sought within this scheme of proof to make out a prima facie case by a combination of static work force statistical evidence covering the charge period, evidence of specific hiring practices followed by ANB during that period, and evidence of specific instances of individual discriminatory actions during the period. In attempted rebuttal of the statistical evidence, ANB sought to establish its “insignificance,” rather than its “inaccuracy” in any computational or objective fact sense. Its chosen means of showing probative insignificance was two-fold: by subjecting it to a standard deviation analysis, to reveal its weakness as proof of a regular operating policy of purposeful discrimination; and by showing, through applicant flow statistical data, that its post-Act decisions were manifestly nondiscriminatory when considered in relation to its hiring opportunities during that period, particularly when subjected to a standard deviation analysis.
It is with respect to the evidence offered by the parties to support these positions that the district judge committed the specific legal errors of analysis that we now address.
First off, the court’s analysis reveals a basic misapprehension of the relationship between statistical and nonstatistical evidence offered to establish a prima facie case, of the relationship between these and evidence tending to rebut any prima facie case established, and of the appropriate mode of analysis under the disparate treatment proof scheme. This appears in the following way.
Although the court concluded that the EEOC had proved a prima facie case of a pattern of racial discrimination in hiring by ANB between 1969 and 1975, 21 F.E.P. Cases at 1552, it then found that “[t]he available hiring statistics, [a standard deviation] analysis, the evidence concerning the hiring practices of defendant, and the 31 *1189cases of unsuccessful black applicants who testified at trial substantially outweigh and refute the EEOC’s statistical evidence.” 21 F.E.P. Cases at 1584. At this point the court moved on to an independent evaluation of the EEOC’s nonstatistical evidence, finding no suggestion of discrimination in either the hiring practices or in the individual claims of discrimination. In this manner the EEOC’s nonstatistical and statistical evidence were required independently to show discrimination sufficient to establish a prima facie case.
While, as indicated, the suggested disparate treatment proof scheme is not ironclad and rigid, the mode of analysis used by the district court so completely skews its substantive underpinnings that the resulting conclusion of a failure of proof is simply not supportable. Under a proper analysis, all of the evidence, statistical and nonstatistical, tending to establish a prima facie case should first have been assessed on a cumulative basis. If that assessment showed a prima facie case made out, inquiry should then have turned to whether, by any of the suggested means, the prima facie case — the inference of discrimination — had been effectively dispelled. Under such an analysis, and with the statistical data correctly assessed, we conclude, for reasons that follow, that on the evidence before the district court, the EEOC with one exception did prove a pattern or practice of discrimination in the respects charged.
A.
We commence in agreement with the district court’s specific conclusion that looking alone to the EEOC’s static work force statistics,7 a prima facie case of discriminatory hiring patterns or practices was made out. As frequently observed by the Supreme Court, and as recognized by the district court, gross statistical disparities in the static work force during the relevant period may alone constitute prima facie proof of the discriminatory practice. Hazelwood, 433 U.S. at 307-08, 97 S.Ct. at 2741-42; see Teamsters, 431 U.S. at 335 n.15, 339 n.20, 97 S.Ct. at 1854 n.15, 1856 n.20; Arlington Heights v. Metropolitan Housing Development Corp., 429 U.S. 252, 265-66, 97 S.Ct. 555, 563, 50 L.Ed.2d 450 (1976). The evidence supporting this fundamental, threshold conclusion by the district court bears emphasis at this point in view of that court’s subsequent conclusion that its inferential force was completely dispelled by other evidence.
The statistical data in respect of officers and managers, using qualified labor pool figures,8 showed that in Suffolk there were no black officials or managers during any of the years in question,9 while the availa*1190ble work force was 8.0 — 10.9% black. In the office and clerical workers category, during three of the charge years (1969, 1974, 1975) there were no blacks employed in Suffolk in these categories. In two years (1970, 1973), there was one black, and in two years (1971, 1972) there were two. When two blacks were employed, the percentage (10.0-11.1%) is equivalent to the percentage of blacks available in Suffolk (10.3%) but far short of the percentage in Nansemond County (22.5%). Further, for five of seven years, 100% of the service workers in Suffolk were black, compared with 44.8-58.1% in the available work force.
The Portsmouth figures are equally compelling. For four of seven years there were no black officials and managers; for the remaining three years there was one black employed in that category (2.7-2.8%). The percentage of blacks in the available work force was 4.8-6.9%. In the office and clerical workers category, the percentage of blacks ranged from 0.0-6.5% for six of the years, with a high of 9.3% in 1975, while the available work force was 13.9-21.5% black. As in Suffolk, the service workers were 80.0-100.0% black while the available work force was only 45.3-59.2% black.
These statistical disparities are substantial, in some cases reaching the “inexorable zero” point. Teamsters, 431 U.S. at 342 n.23, 97 S.Ct. at 1858 n.23. They show that blacks were consistently underrepresented in the office and clerical categories in branches in both cities and unrepresented in the officials and managers categories in Suffolk for all years and Portsmouth for four of seven years. The district court’s conclusion that, considered alone, they establish a prima facie case is firmly supported by the record.
We turn now in detail to the process of analysis by which, starting from this point, the district court ultimately determined that this prima facie statistical case was defeated. One neutralizing factor for the district court was the perceived effect of a standard deviation analysis10 upon this *1191static work force statistical data. In the officer/manager category the court found that the standard deviations always remained less than two, and in the office and clerical employees category, the standard deviations were always less than three for Suffolk and almost always more than three for Portsmouth. This analysis, according to the district court, neutralized the prima fa-cie statistical case in all but the office and clerical workers category for Portsmouth.
*1190
*1191We disagree with this conclusion derived by the district court from its standard deviation analysis — for reasons that we think important to spell out. The conclusion was based upon an apparent assumption that if standard deviations reflected in static work force statistics were not “more than two or three” the disparities were necessarily shown to be statistically insignificant. 21 F.E.P. Cases at 1558-59. This assumption is simply incorrect — for reasons we think it important to develop in order to guard against misuse of this method of analyzing statistical proof and to emphasize its limitations when used by courts in this type case.
The district court’s assumption was presumably drawn from general observations made by the Supreme Court in the course of footnote discussions of standard deviation analysis as a means of testing statistical proof in Castaneda v. Partida, 430 U.S. 482, 496 n.17, 97 S.Ct. 1272, 1281 n.17, 51 L.Ed.2d 498 (1977) and again in Hazelwood, 433 U.S. at 311 n.17, 97 S.Ct. at 2743 n.17. As briefly explained and applied in those opinions, standard deviation analysis may perform some legitimate service for courts in assessing the statistical significance of data offered to establish (Castaneda) or rebut (Hazelwood) a charge of discrimination in composing juries (Castaneda) or work forces (Hazelwood). Simply put, it tests the hypothesis that underrepre-sentation of a protected minority group in any sample made up of a protected and a nonprotected group (binomial distribution) might be attributable to normal fluctuations of chance rather than to discriminatory design. The “standard deviation” is the measure of the predictable fluctuation in a random selection process. The difference between actual (“observed”) numbers of the protected group in such a sample and the number that would be “expected” in a perfectly proportional process of selection from the appropriate pool can then be expressed in numbers of standard deviations. In turn, standard deviations can be expressed in terms of the mathematical probability that chance is the cause of the disparities (differences between “observed numbers and expected values”) measured. As standard deviations increase numerically, the probability of chance as the cause of revealed underrepresentation of course diminishes. To the extent the probability of chance is shown to be quite small, the legal inference of discrimination based upon a rough legal assessment that disparities are manifestly “gross” or “substantial” is thus “scientifically” confirmed.
*1192It was against this general background that the Supreme Court in Castaneda noted that “[a]s a general rule . . . ., if the difference between the expected value and the observed number is greater than two or three standard deviations, then the hypothesis that the [selection process] was random would be suspect to a social scientist.” 430 U.S. at 497 n.17, 97 S.Ct. at 1281 n.17. The Court then pointed out that the standard deviations revealed in the data before it were so far beyond this “suspect” range— 29 for one sample, 12 for another — that the probability of chance as the explanation was simply infinitesimal: in the convenient mathematical short-forms — 1 in 10140 and 1 in 1025, respectively. In this context, it is obvious that the Court was merely emphasizing the lack of any theoretical possibility that chance rather than discriminatory design underlay the underrepresentation it was considering.
If a legal rule of analysis can properly be derived from the Castaneda footnote, it can only be that standard deviations greater than two or three necessarily exclude chance as a cause of underrepresentation. The converse of this — that standard deviations of not “more than two or three” necessarily exclude discriminatory design as the cause — is nowhere implied. Nor could it be, as we shall now attempt to show.
The Castaneda Court had no need to ex-plpre the levels of probability that exist in the range of “two or three” standard deviations, being content to note that beyond this range social scientists would find “suspect” for scientific purposes the hypothesis of random choice. When this range is explored, however, it appears that well short of three standard deviations the probability levels for chance as explanation have already dropped far below the point at which courts of law — concerned with proof by the “greater weight” or “preponderance” of the evidence — would presumably have discarded the hypothesis of chance. Just short of two standard deviations — specifically at 1.96 — the probability of chance is only 5 in 100; at just over two and one half, it is only 1 in 100; by three it is less than 1 in 100. W. Hays & R. Winkler, Statistics: Probability, Inference and Decision 218-19, 381-82 (1971). For this reason, authority can be found for the proposition that most social scientists, applying laboratory rigor to rule out chance as even a theoretical possibility rather than the law’s rougher gauge of the “preponderance of the evidence,” are prepared to discard chance as an hypothesis when its probability level is no more than 5%, i. e. at approximately two standard deviations. Id. at 394.
From all this we conclude that courts of law should be extremely cautious in drawing any conclusions from standard deviations in the range of one to three. Above this range, with standard deviations of more than three, the analysis may perhaps safely be used absolutely to exclude chance as a hypothesis, hence absolutely to confirm the legitimacy of an inference of discrimination based upon judicial appraisals that disparities are, to the legally trained eye, “gross.” This we conclude is all that the Supreme Court has ever directly approved by its own use of the process.11 Within the range of one to three standard deviations, where the probability of chance *1193as explanation for revealed underrepresen-tation declines precipitately from only 5% at two standard deviations to less than 1% at three, we do not see how a court can properly find the only other hypothesis— discrimination — dispelled by this analysis alone. On this basis, we conclude that the district court’s analysis revealing standard deviations which, as applied to the aggregate of years and work categories involved, lay preponderate^ within or above this range, could not properly be taken as dispelling the inference of discrimination based upon the disparities in the static work force data. To the extent this was the basis of the district court’s finding, we conclude that it was clearly erroneous.12
B.
We turn now to the district court’s assessment of the defendant’s applicant flow data13 offered to rebut the prima facie case of discrimination made out by the static work force statistics. ANB contends that it does so in two approved ways: first, by affirmatively showing that ANB’s hiring decisions during the charge periods were not discriminatory; alternatively by demonstrating that the total number of hiring decisions during the period were so few in total number that, without regard to their substance, they could not support any conclusion of a discriminatory pattern or practice underlying them. As earlier noted in this opinion, both of these avenues of rebuttal are open to an employer, and the district court apparently relied upon them as alternative grounds in concluding that the prima facie case had been rebutted. See note 6 supra.
We address these in reverse order, looking first to the conclusion that the number of hiring decisions was insufficient to support a conclusion of discrimination in their making. Because the prima facie statistical proof of discrimination was keyed to different employment categories based upon qualifications, hence involved different labor pools for base data, analysis of the sufficiency of rebuttal proof in these two respects must similarly be separately assessed.
ANB’s rebuttal applicant flow evidence showed that in Suffolk ANB hired 35 clerical employees and 2 managers over the seven year period 1969-1975, while in Portsmouth during the single year 1975 it hired a total of 23 clerical employees and 2 managers. 21 F.E.P. Cases at 1556 n.51. These are, in absolute terms, concededly small samples from which to attempt straight comparisons with applicant pools or general labor pools as sources of base data. Samples too small are suspect as a basis upon which to infer any pattern of discrimination in making employment decisions. See May- or of Philadelphia v. Educational Equality League, 415 U.S. 605, 620, 94 S.Ct. 1323, 1333, 39 L.Ed.2d 630 (1974). But, aside from the Supreme Court’s general observa*1194tion in Mayor oí Philadelphia that on the very special facts there presented “the District Court’s concern for the smallness of the sample presented by the 13-member Panel was also well founded,” id. at 621, we have no guidance as to how small is too small other than our own best judgment.
There are two conflicting considerations to be kept in mind in exercising that judgment. The danger of unfairness to the employer in resting inferences of discriminatory employment practices on proof involving small total members of employment decisions is obvious. But there is the countervailing consideration that, given the difficulties of proving discriminatory motive under any circumstances, see Brown v. Gaston County Dyeing Machine Co., 457 F.2d 1377, 1382 (4th Cir. 1972); United States v. Jacksonville Terminal Co., 451 F.2d 418, 442 (5th Cir. 1971), a too ready rejection of claims solely on this account practically precludes proof of discrimination in circumstances involving local employers with relatively small total work forces. Courts have simply to balance the two with an eye to protecting against purely speculative findings of discrimination while not cutting off the claims of some employees simply because of the small overall size of the work forces in which they happen to be employed. See Chicano Police Officers’ Ass’n v. Stover, 526 F.2d 431, 439 (10th Cir. 1975). This difficult balancing problem is made even more difficult where, as here, the charged discrimination runs to several categories of differently qualified employees so that, under developed doctrine, proof must be assessed separately as to each. In this situation we think it is entirely proper in gauging the danger of unfair inferences from small numbers in respect of one category to take into account — for this limited purpose — any patterns inferable from the total range of hiring decisions affecting all categories during a charged period of discrimination.
Guided by these general considerations, we cannot find error in the district court’s conclusion that the total number of hiring decisions affecting managers in the Suffolk branches — on ANB’s undisputed evidence but two over a seven year period— was too small fairly to support any inference of a discriminatory pattern of hiring. On this basis we agree that the prima facie case was rebutted with respect to this category of employee notwithstanding the proof pro and con with respect to other categories.
With respect to the numbers of decisions affecting both managers and office/clerical employees in the Portsmouth branches, we find error in the district court’s apparent conclusion that those numbers were also insufficient for inferential purposes. They were limited in ANB’s proof to those for just one year, 1975. For reasons more fully developed in our later discussion of the sufficiency of the applicant flow data affirmatively to rebut the prima facie case, we do not consider this omission of proof running to the whole record excusable on ANB’s part. Accordingly, we conclude that as to these categories, the prima facie statistical case is not rebutted by this obviously incomplete showing of the total number of hiring decisions made over the charged period.
With respect to the number of hiring decisions affecting office/clerical positions in Suffolk over the charged period— thirty-five on ANB’s proof — we think it quite sufficient as a basis for inferring the pattern of discrimination prima facie established. Accordingly, we find error in the district court’s contrary conclusion as to this category.
More fundamentally, the district court concluded that ANB’s applicant flow data in any event sufficed affirmatively to rebut the statistically based prima facie case by showing that its hiring decisions during the charge period were not discriminatorily motivated. Hazelwood has of course established that a prima facie statistical case can be so rebutted. 433 U.S. at 313, 97 S.Ct. at 2744. Because employment discrimination was not illegal under Title VII until 1965, “[an] employer who from that date forward made all its employment decisions in a wholly nondiscriminatory way would not *1195violate Title VII even if it had formerly maintained an all-white work force by purposefully excluding Negroes.” Id. at 309, 97 S.Ct. at 2742. Applicant flow data obviously is adaptable to that end. To be considered for this purpose, however, the particular data offered must be “sufficiently reliable ... to permit consideration of [an employer’s] argument that those data may undercut a statistical analysis dependent upon hirings alone.” Id. at 313 n.21, 97 S.Ct. at 2744 n.21.
Over the EEOC’s objections the district court found ANB’s applicant flow data sufficiently reliable to consider in rebuttal of the work force statistics. When the court then assessed the data it concluded that because it did not give rise to an inference of discrimination, it sufficed to rebut the prima facie case made out by the static work force statistics. 21 F.E.P. Cases at 1555-56. We think the district court erred in its conclusion that the data was sufficiently reliable to be considered as rebutting evidence. Its reliability is suspect on three separate bases which in conjunction undercut its probative force for the intended purpose.14
First, the applicant flow data for the Portsmouth branches is limited to only one of the charged years, 1975. Applicant flow data limited to one out of seven relevant years cannot be held to rebut a prima facie case based upon gross disparities revealed in static work force statistics over the period. It simply has not that probative force. The district court recognized its unreliability because of incompleteness, but excused the incompleteness because of what it considered the EEOC’s bad faith in failing to notify ANB until commencing action that the Portsmouth branches were to be included in the charges.
In consequence ANB had destroyed its pre-1975 records for Portsmouth, a decision which the district court concluded was reasonable and which accordingly should not be allowed to prejudice ANB in presenting its rebuttal evidence. We find nothing in Title VII, EEOC regulations, or Supreme Court interpretations of Title VII that would operate to relieve employers of the normal consequences of such a deficiency of proof.
The district court found sufficient excuse by negative implication from the EEOC’s requirement that all applications for employment be routinely preserved for six months, and that after a charge has been filed all relevant records be maintained until final disposition. 29 C.F.R. § 1602.14(a) (1976). From this the court concluded that ANB was reasonably entitled to decide, after the reasonable cause determination which did not include Portsmouth had been made, that preservation of relevant application forms for those branches were not necessary. 21 F.E.P. Cases at 1555.
Leaving aside all questions of the reasonableness of such a business decision, we disagree with the consequence given it by the district court. The affirmative obligation imposed by § 1602.14(a) to preserve records was clearly designed to protect Title VII plaintiffs from an employer’s destruction of possibly damaging evidence. This being its purpose, it cannot sensibly be interpreted as being intended also to protect employers against the consequences of their voluntary destruction of such records just because the affirmative obligation to preserve them has expired. So to interpret it would license a self-serving destruction of records by avoiding the normal factual inference of self-serving that arises from the destruction of evidence. This we think could not accord with the broad remedial purposes of Title VII. See EEOC v. Cook Paint & Varnish Co., 24 F.E.P. Cases 51, 55 (W.D.Mo.1980). Employers have been on notice since the earliest days of Title VII’s enforcement of the critical importance of *1196the maintenance of employment records going back at least to the effective date of the Title. In consequence, holding this employer to the normal litigation consequences of a failure to maintain relevant employment records imposes no higher standard than that dictated by sound business judgment in respect of the maintenance of all business records having potential relevance in any of the litigation patterns to which businesses stand constantly exposed.
Next, the data’s reliability is open to serious question because it apparently reflected less than half — the EEOC contended only 46% — of the actual applicants from 1969-1975. The district court dismissed the EEOC’s contention of its unreliability on this score, finding that the contention was based upon a confused reference to the proportion of the available applications requested by the EEOC during discovery. 21 F.E.P. Cases at 1554. Whether or not the EEOC confused its references, it seems highly unlikely on the record we review that this data represents substantially all the applicant flow data for the years in question. As the district court noted, the total number of applications reflected in the data offered the court varied widely from year to year, from a low of sixteen in 1972 to a high of ninety-seven in 1974. Id. at 1554 n.42.
Finally, the data considered by the district court included service workers.15 Because 100% of the service workers hired in both Suffolk and Portsmouth were black, including them distorted the picture of minority hiring in the contested categories. If we limit the applicant flow data to office/clerical workers categories, purging the data, as offered, of all officials/managers and service worker hires,16 it completely *1197fails to dispel the inference of discrimination found by the district court to have been created by the EEOC’s statistical proof. In Suffolk, ANB never hired more than one black clerical worker at a time — in some years none. The Suffolk samples for each year are concededly small, but the overall results simply confirm rather than dispel the prima facie case based upon static work force statistics: over these years, during which blacks made up 24.8% of the qualified applicant pool, only 14.2% of those hired in these categories were black. In Portsmouth the limited data for 1975, purged of service workers, not only fails to dispel the prima facie case but reinforces it. ANB’s seven Portsmouth branches in 1975 hired one black clerical worker — 4.3% of its hires — out of an applicant pool 23.1% black.17
*1196Suffolk Branches
Suffolk Branches
These figures translate into the following percentages:
Percentage of
Portsmouth Branches
*1197We conclude, with respect to the Office/Clerical category in Suffolk and Office/Clerieal and Officials/Managers categories in Portsmouth, that the district court erred in ascribing to the defendant’s applicant flow data the neutralizing effect reflected in that court’s decision. At this stage in its analysis, the district court should instead have concluded, as we now do, that the prima facie case of discrimination for those categories made out on the static work force statistics had not been rebutted by this data.
C.
We turn next to the way in which the district court assessed the EEOC’s nonsta-tistical evidence of specific hiring practices. Specifically the EEOC alleged and introduced evidence to prove that between 1969 and 1975 ANB failed to use objective, job related standards, used inconsistent standards and perpetuated the predominantly white work force through a preference for friends and relatives of employees and word-of-mouth recruiting (including walk-in hiring), and by maintaining an all-white interviewer staff.
The district court concluded that all these practices, individually and collectively assessed, were lawful and created no inference of discrimination.18 Specifically, the court concluded that hiring was not con*1198ducted primarily through “word of mouth recruiting”; that there was no preference for friends or relatives of employees; that most of the hiring standards used, though unwritten, were sufficiently specific and objective, and were consistently applied; that the subjective evaluations of attractiveness, neatness, maturity and personality, and ability to communicate were justified by legitimate and important business considerations and applied in a nondiscriminatory way; and that the use of an all-white interviewing staff had no discriminatory effects. From this the court concluded that the hiring practices evidence actually operated with the applicant flow data and the standard deviation analysis to undercut the statistically based prima facie case, rather than to corroborate or bolster it as intended by the EEOC. In this conclusion we find error of law.
While we are not prepared to say that attempted proof of this sort may never, by reason of its intrinsic weakness, have such a negative effect upon a prima facie statistical case, this evidence could not properly be given such an effect. The district court’s legal analysis leading to its contrary conclusion is flawed in two ways.
First, it failed to assess the evidence in the light of, and colored by, the gross underrepresentation of blacks in ANB’s work force already statistically demonstrated to the district court’s satisfaction. This was at odds with the teaching of such cases as Barnett v. W.T. Grant Co., 518 F.2d 543 (4th Cir. 1975) and Rock v. Norfolk & Western Ry., 473 F.2d 1344 (4th Cir. 1973), that where such an imbalance in the work force has been statistically shown, the uses of particular hiring practices are then to be assessed for their tendency to perpetuate that imbalance, an assessment in which they may properly be found to be “badges of discrimination that serve to corroborate, not to rebut, the racial bias pictured by the statistical pattern of the ... work force.” Brown v. Gaston County Dyeing Machine Co., 457 F.2d 1377, 1383 (4th Cir. 1972). Here, rather than analyzing the hiring practices on a basis which assessed their tendency to perpetuate an existing condition of underrepresentation, the district court approached them as if no such imbalance existed. Inquiry was confined to specific practices isolated from any inferences of discrimination already established. It was, in effect, as if the EEOC had sought to base its case solely upon the nonstatistical evidence related to hiring practices. The resulting analysis failed properly to take into account the backdrop of existing imbalance against which the practices were carried out and in consequence misapplied controlling legal principles to the evidence.
This general approach obviously colored the district court’s overall analysis of the specific hiring practices challenged by the EEOC. In addition, we think the assessment of certain of the specific practices was further tainted by faulty analysis of the evidence addressed to those practices. To show why, we briefly summarize the pertinent evidence and the district court’s assessment.
The uncontradicted evidence showed that ANB relied almost exclusively upon walk-ins as a primary source of new employees, never advertising a vacancy unless it had exhausted the file without filling the position. Applicants were required to list on their applications friends or relatives employed by the bank. Thirty-two of sixty-five applicants hired in Suffolk from 1969-1975 and in Portsmouth in 1975-49.2% — had listed friends or relatives on their applications. Further, the applicants were screened and interviewed by an all-white interviewer staff whose evaluations of their attractiveness and neatness were very important factors in hiring. The EEOC contended that this evidence was probative of one of the traditional “badges of discrimination”: word-of-mouth recruiting rather than general advertising through established public channels to fill vacancies. Specifically, this was said to be established as the standard hiring practice by the facts that (1) friends and relatives of employees were demonstrably preferred, (2) vacancies were not advertised until after on-file applications were depleted, and (3) no notices of *1199vacancies were posted to employees. The court essentially rejected this evidence as having no probative force. It did not show that word-of-mouth was the primary hiring technique of the defendant because (1) there was no preference for friends or relatives; (2) ANB “did attempt to ‘advertise’ itself to qualified minorities outside its work force”;19 (3) the screening process was not “tainted with discriminatory intent [nor did it operate] (through word-of-mouth recruitment or preference for friends or relatives of employees, for example) to perpetuate minority representation”; and (4) the failure to post notices to its employees was not a violation of Title VII because the evidence of discrimination was at best inconclusive. 21 F.E.P. Cases at 1566.
The court’s finding that there was no preference for friends and relatives was crucial to the conclusion that word-of-mouth recruiting was not shown to be the primary means for filling vacancies. The evidence showed that thirty-two of the sixty-five hires — 49.2%—at Suffolk in 1969-1975 and Portsmouth in 1975 had listed friends or relatives on their applications. From this the court concluded that “[i]f anything, the . . . statistics show[ed] that the listing of friends and relatives was a minor factor; a majority of the hires had no ‘contacts’ working for defendants.” Id. at 1565.
Without ascribing controlling significance to this particular factor in the overall assessment of ANB’s hiring practices, we observe that if anything, this data implies that the listing of friends and relatives did indeed tend to perpetuate the underrepre-sentation of blacks in non-service jobs and their concentration in service jobs.20 Whether this practice was a major factor in hiring or not, its unmistakable tendency to perpetuate the existing racial imbalance in the work force could not properly be dismissed as of no consequence.
In similar fashion, the district court essentially rejected out of hand any significance for the fact that throughout the charged period, ANB used an all-white interviewing staff. We cannot quarrel with the court’s related conclusions that the criteria used for selection were for the most part objective and consistently applied, and that the subjective evaluations of attractiveness and neatness were business-justi*1200fied. Furthermore, it is clear that the use of an all-white interviewing staff standing alone could not súpport a determination of liability. Nevertheless, it was error to reject as having no probative force on the issue of discriminatory hiring practices the fact that in a racially imbalanced setting, the staff charged with the duty of evaluating personal characteristics of job applicants was maintained all-white throughout the charged period. See Fisher v. Procter & Gamble Manufacturing Co., 613 F.2d 527, 545 — 46 (5th Cir. 1980); Domingo v. New England Fish Co., 16 E.P.D. 18207 at 5070, 5080 (W.D.Wash.1977).
*1199
*1200In summary, on this aspect of the case we conclude that, rightly assessed, the EEOC’s evidence of word-of-mouth recruiting as the primary means used by ANB to fill vacancies, and the use of an all-white interviewing staff to make its subjective hiring evaluations tended to corroborate— to some degree at least — the prima facie showing of discrimination made by the static work force statistics. The district court’s apparent conclusion that it had no such probative force or that its lack of force tended indeed to rebut the prima facie statistical showing reveals a misapprehension or misapplication of controlling principles of law.
D.
We now briefly consider the district court’s analysis of the EEOC’s evidence of individual instances of discrimination. As earlier indicated, 31 of 52 black persons identified by the EEOC as victims of individual acts of discrimination in hiring gave testimony, and the district court found on the evidence presented that none was a specific victim of discrimination. In consequence, as with the hiring practice evidence, the district court dismissed this evidence as having no probative force in respect of the pattern or practice issue, or possibly treated it as having a negative impact for that purpose.
Because we conclude that the EEOC’s prima facie case was otherwise established and not rebutted without reference to this evidence, we find it unnecessary to review the district court’s findings and conclusions which led to its out of hand rejection. We observe only that the evidence was not offered at this stage to establish entitlement to individual relief, but merely as corroborating or buttressing evidence of a general pattern of racial discrimination in hiring. Assuming for purposes of this appeal that its cumulative impact for this purpose was rightly assessed by the district court as not buttressing the EEOC’s other evidence, we hold that neither could its cumulative impact be properly adjudged to rebut the EEOC’s case. Given the basis of our disposition, it may simply be set aside as a neutral factor in the overall assessment of the evidence.
E.
Looking now to the whole body of evidence before the district court, we hold, in concluding summary, that (1) the district court correctly determined that the EEOC’s static work force statistics established a prima facie case of discriminatory pattern or practice of hiring during the charged period; (2) except with respect to the Officials/Managers category in Suffolk, the district court erred as a matter of law in holding that this prima facie case was rebutted by the combined force of the standard deviation analysis applied to the static work force statistics, the applicant flow data for the charged period as tested by a standard deviation analysis, and the weakness or negative impact of the EEOC’s non-statistical evidence of hiring practices and individual instances of discrimination. Analyzed free of the identified errors of misapprehension and misapplication of controlling legal principle, we conclude that the evidence before the district court established a prima facie case based upon the statistically revealed gross disparities in the work force; that this was buttressed to some degree by evidence of hiring practices which tended to perpetuate the undisputed racial imbalance; and that this prima facie case — except in one respect — was never rebutted by legally sufficient evidence.
*1201Accordingly, we hold that, except in respect of the one category of officials/managers in the Suffolk branches, the total evidence established the existence of a discriminatory pattern or practice of hiring in violation of Title VII that requires remand for the determination of appropriate relief.
IV
Upon remand the EEOC is entitled to án injunctive decree . appropriately shaped to remedy the consequences of the discriminatory hiring policy we find conclusively established upon the record, and to further proceedings in which individual persons claiming to have suffered the consequences of that pattern or practice during the charged period may have their individual claims adjudicated. In these Stage II proceedings, an individual claimant is entitled, upon proving only that he or she was a black who applied unsuccessfully for a job with the defendant during the relevant period, Sledge v. J.P. Stevens & Co., 585 F.2d 625, 637 (4th Cir. 1978), to the benefit of an inference that the decision not to hire “was made in pursuit of that policy [of discrimination].” Teamsters, 431 U.S. at 362, 97 S.Ct. at 1868. The burden then shifts to the employer to show “that the individual applicant was denied an employment opportunity for lawful reasons.” Id. This burden involves showing by a preponderance of the evidence that nondiscriminatory factors motivated the decision not to hire. Sledge v. J.P. Stevens, 585 F.2d at 637. If the defendant meets this burden, the individual claimant is then entitled to show that the “exculpatory reason advanced by the employer is mere pretense — that, for example, white persons with qualifications identical to those of the claimant or who also did not meet the prescribed qualifications were nonetheless assigned to the position sought.” Id.
As earlier noted, the district court has made findings and conclusions in respect of some persons who may present claims in the Stage II proceedings. Those findings and conclusions, made in connection with the general issue of the existence of a discriminatory pattern or policy and under different burdens of proof, have no preclusive effect in Stage II proceedings.
Consistent with the result we reach on the merits, we vacate the district court’s award of attorney’s fees in favor of ANB.
AFFIRMED IN PART; REVERSED IN PART; AND REMANDED.
APPENDIX A
*1202
APPENDIX B
*1203
*1204Combined Figures (Suffolk and Portsmouth Branches) Officials and Managers Number and Percentage of Blacks 1968 1969 1970 1971 1972 1973 1974 1975 Number 0 0 0 0 0 1 1 1 Percentage 0.0 0.0 0.0 0.0 0.0 2.3 2.4 2.3 Percentage of Blacks in Available Work Force Suffolk Nansemond Portsmouth 8.0 8.0 10.9 10.9 10.9 10.9 10.9 10.9 10.9 10.9 6.9 6.9 6.9 6.9 6.9 6.9 6.9 6.9 Norfolk-Portsmouth SMSA 4.8 4.8 4.8 4.8 4.8 4.8 4.8 4.8
APPENDIX C
*1205
Portsmouth Branches
These figures translate to the following percentages:
. The EEOC offered two different definitions of the relevant market area; ANB neither challenged the EEOC’s approaches nor offered its own definition. The district court found the relevant labor market area for ANB’s Portsmouth branches to be the City of Portsmouth and the Norfolk-Portsmouth SMSA; for the Suffolk branches, it defined the relevant area as the city of Suffolk and Nansemond County.
The court used the data set out in Appendix A for general work force comparisons for the years 1968-1975, comparing for each year the number and percentage of blacks employed by ANB with the percentage of blacks in the relevant market area.
The data employed here are of course subject to the general limitation implicit in the term “static.” They only depict the work force as it existed on specific days at intervals of a year during the charged period. They do not therefore purport to reflect all employment decisions occurring during those intervals, and may “miss” significant hirings occurring during those intervals.
It is this general limitation rather than any specific inaccuracy in the data here used by the district court that the dissent has pointed up. Slip op. at 93-95. While this limitation can of course give a distorted picture on the ultimate issue of post-Act discrimination, the means for correcting it are provided by the same Supreme Court decisions that authorize use of static work force statistics to make a prima facie case. See generally note 7 infra.
. The court used the data set out in Appendix B for special qualifications work force comparisons for the years 1968-1975, comparing for each year the number and percentage of blacks employed by the defendant in Suffolk and Portsmouth in each category with the percentage of qualified blacks in the available work force.
. Hiring procedures were similar at the Portsmouth and Suffolk branches.
. See Appendix C.
. The situation is similar to one that occasionally occurs in the private suit where the plaintiff names a defendant not named in the original charge filed with the EEOC. The general rule is that “[i]t is a jurisdictional prerequisite to the filing of a suit under Title VII that a charge be filed with the EEOC against the party sought to be sued.” Bowe v. Colgate-Palmolive Co., 416 F.2d 711, 719 (7th Cir. 1969); see Stith v. Manor Baking Co., 418 F.Supp. 150, 155-56 (W.D.Mo.1976), and cases cited therein. The rationale for the rule is like that behind the rule that the claims in a EEOC enforcement suit must have been included in the reasonable cause determination and subject to conciliation. The defendant, if he has been named in the original charge, has been notified of the asserted Title VII violation. Further, the charged party is thus brought before the EEOC, “[permitting] effectuation of the Act’s primary goal, the securing of voluntary compliance with the law.” Bowe v. Colgate-Palmolive Co., 416 F.2d at 719. Courts have developed exceptions to this rule, though, where it is clear that the defendant through some relationship with the named respondent had notice of the charges and participated in the conciliation process. See, e. g., Stith v. Manor Baking Co., 418 F.Supp. at 156, and cases cited therein; Escamilla v. Mosher Steel Co., 386 F.Supp. 101, 105 (S.D.Tex.1975) (jurisdiction proper over parent of wholly-owned subsidiary where parent had or should have had notice of conciliation process); Chastang v. Flynn & Emrich Co., 365 F.Supp. 957, 964 (D.Md.1973) (“where there is substantial, if not complete identity of parties before the EEOC and the court, it would require an unnecessarily technical and restrictive reading of [the statute]” to deny jurisdiction), aff'd in relevant part, 541 F.2d 1040 (4th Cir. 1976).
. The district court apparently concluded alternatively that, applying this principle, ANB’s hiring decisions during the relevant period, as revealed in its applicant flow data, were too few in number (39 in Suffolk from 1968 to 1975 and 27 in Portsmouth in 1975) to justify an inference of discrimination. 21 F.E.P. Cases at 1556.
. Section 703(j) of Title VII “imposes no requirement that a work force mirror the general population.” Teamsters v. United States, 431 U.S. 324, 339 n.20, 97 S.Ct. 1843, 1856 n.20 (1977). Employment discrimination is seldom overt, however, and courts must look to evidence other than specific acts of discrimination to determine whether discrimination exists. Brown v. Gaston County Dyeing Machine Co., 457 F.2d 1377, 1382 (4th Cir. 1972). Static work force statistics showing a racial imbalance are probative in these cases not because the imbalance shown is in itself violative of Title VII, but because “such 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. Evidence of longlasting and gross disparity between the composition of a work force and that of the general population thus may be significant....” Teamsters, 431 U.S. at 339 n.20, 97 S.Ct. at 1856 n.20.
. The district court analyzed, as well, general work force figures. See Appendix A. We recognize that in EEOC v. United Virginia Bank/Seaboard National, 615 F.2d 147, 150 (4th Cir. 1980), we upheld under Fed.R.Civ.P. 52(a) a finding by the district court that general labor force statistics were not an appropriate statistical group for comparison with bank employees. Without further analysis we here rely only on the specialized work force figures, which clearly support an inference that race was a factor in hiring. See Appendix B for this data.
. The district court included 1968 in its charts, but the charge years included only 1969-1975. The practices apparent in 1968 are consistent with the pattern in later years. Even so, in finding an inference of discrimination we have considered only 1969-1975.
. The district court applied the standard deviation analysis to the specialized work force figures, recording generalized values for the number of standard deviations revealed, as follows:
. In Hazelwood the Court used a standard deviation analysis to emphasize the importance of choosing the right labor pool from which to derive base data for showing statistical disparities. The Court was careful to emphasize that this was the sole purpose and that its “observations were not intended to suggest that precise calculations of statistical significance are necessary in employing statistical proof .... ” 433 U.S. at 311 n.17, 97 S.Ct. at 2743 n.17. Noting, in paraphrase of the above-cited passage from Castaneda, that “a fluctuation of more than two or three standard deviations would undercut the hypothesis that decisions were being made randomly with respect to race,” id., the Court then proceeded to point out that depending upon which of two possible labor pools were chosen, the standard deviations could range from less than one to more than six. Significantly, the only direct consequence for proof assessment noted by the Court was that standard deviations of more than two or three would “undercut” the hypothesis of random selection. To the extent there is any implication that the low-side deviations might rebut, it may be significant that they were all less than two.
. Another factor suggesting great caution in making fine-tuned use of standard deviation analysis in these cases is that its reliability diminishes in ways probably not susceptible of precise handling by courts as the binomial distribution sample diminishes in size. See W. Hays & R. Winkler, Statistics: Probability, Inference and Decision 222 — 26 (1971). The Supreme Court has not indicated the size samples reliably susceptible to this analysis. Castaneda, dealing with an observed figure of more than 300 in a sample of 870, referred in its discussion to the general rule “for such large samples.” 430 U.S. at 496 n.17, 97 S.Ct. at 1281 n.10. The smallest sample in Hazelwood where the same sort of standard deviation analysis was discussed was 123. 433 U.S. at 311 n.17, 97 S.Ct. at 2743 n.17. This is therefore the smallest sample considered by the Court in this context, as noted by the district court. 21 F.E.P. Cases at 1558 n.54. Significantly, the categories in the district court’s analysis which yielded the smallest standard deviations — Suffolk official/managerial employees — also represented the smallest samples: four to five. We are aware that in EEOC v. United Virginia Bank/Seaboard National, 615 F.2d 147 (4th Cir. 1980), a divided panel of this court used a standard deviation analysis in respect of samples even smaller than some in the present case. That panel recognized, however, that at some point sample size affects the reliability of this mode of analysis, id. at 151, 152, and employed it there as merely one of several factors demonstrating error in a district court’s factual determination that a statistically based prima facie case had been established.
. See Appendix C.
. We do not consider still another possible basis of unreliability because of a lack of factual basis for assessing it. In some cases the comparison of hires to the relevant applicant pool for rebuttal may be entirely inappropriate because the employer’s discriminatory hiring practices are so well known throughout the community that blacks may be reluctant to apply because such an effort would be futile. See, e. g., Lea v. Cone Mills Corp., 301 F.Supp. 97, 102 (M.D.N.C.1969), aff’d in relevant part 438 F.2d 86 (4th Cir. 1971). There is no direct evidence to support this in the record.
. ANB contests the EEOC’s assertion that the district court “lumped” together managerial, clerical and janitorial categories for its analysis of the applicant flow data. Brief for Appellee 31 n.15. Both parties are correct. The court combined all hires in its general analysis of the statistical disparities, 21 F.E.P. Cases at 1555-56, but separated out janitorial and managerial hires for its standard deviation analysis of the figures. Id. at 1559-60. However, in assessing the correctness of the district court’s general evaluation of the applicant flow data, we note that for that purpose all employees were “lumped” together.
. Th\s clerical category in Suffolk is all we consider here, given our conclusion that the prima facie case with respect to managers had been rebutted. Further, consistent with our position that where special qualifications are required the rebuttal data must also be so categorized, we approach the Portsmouth data by category and consider only the clerical hires due to the small number of managerial hires. The district court noted that two management trainees and two black service workers were hired in Suffolk during 1969-1975 and in Portsmouth in 1975. 21 F.E.P. Cases at 1559 n.57, 1560. Removing these hires and all service-worker applicants from the data, we are left with:
. The district court applied a standard deviation analysis to the office and clerical worker segments of the applicant flow figures, correctly finding the officials and managers categories of hires too small for such analysis. 21 F.E.P. Cases at 1559-60. In doing so, although the statistical and numerical comparisons were between the annual applicant pools and hires, the court analyzed the hires in relation to the qualified labor pool statistics for each city rather than the applicant pool. The court found this comparison mandated by Hazelwood and Castaneda. Id. at 1556 n.53. We find no such rule emanating from either case, noting that Hazel-wood’s discussion was merely collateral given that the Court ordered a remand to the district court for further findings on that data, 433 U.S. at 313, 97 S.Ct. at 2744, and pointing out also that Castaneda’s comparison of jury composition to general population figures was not only logical but unavoidable because juries are not selected from applicants but from the citizenry at large. Further, it is manifestly incorrect to compare hires with the applicant pool and then test these statistical comparisons by a standard deviation analysis based on different population statistics. Therefore, the district court’s particular use of the standard deviation analysis was without legal or factual basis, and we disregard it entirely. Because of the demonstrated unreliability of these figures there is no need to attempt a correct reanalysis of the statistical significance of these disparities.
. The district court also examined a number of specific practices “which the EEOC, before and at trial, suggested were discriminatory.” 21 F.E.P. Cases at 1560. Many of these allegations were made in the reasonable cause determination but were apparently not actively pursued at trial. The district court stated repeatedly in evaluating these claims that the EEOC put on no evidence to support them. We find no error in the resulting findings and conclusions regarding the discriminatory effect of retail credit checks, the requirement for a high school diploma, the requirement for character and employment references, testing, the failure to correspond with applicants or college education as a negative factor. To the extent the EEOC’s failure to follow up charges with attempted proof reflected irresponsible broadside charging — a conclusion apparently drawn by the district court — we obviously do not condone the practice and indeed approve the district court’s implicit condemnation of it. This does not, however, relieve either that court or this one of the obligation to subject evidence actually adduced of other hiring practices to independent, objective evaluation.
. The district court concluded at one point that the only word-of-mouth recruiting shown was that involving some affirmative efforts made by ANB to recruit qualified black employees, and that this obviously did not tend to prove discriminatory motive in hiring. While the court did consider and reject other suggestions of word-of-mouth recruiting, its apparent perception that this was the only direct evidence on the issue in the record clearly influenced the ultimate conclusion that the EEOC had failed to establish word-of-mouth recruiting as its primary hiring procedure. This completely warps the meaning and significance of the term in the context of Title VII litigation. As developed in this context, the term connotes a passive dependence for applicants upon informal advertising of vacancies through employees’ communications to friends, relatives and acquaintances and upon walk-in applicants, rather than upon systematic public advertising designed to reach the qualified available labor pool. There was of course substantial direct evidence in the record of word-of-mouth recruiting so understood.
Significantly, the court did not find and ANB does not contend that the affirmative action recruiting effort upon which the court focused in assessing the claim of word-of-mouth recruiting produced any results. Properly assessed this evidence simply shows one episodic means of recruiting used by ANB within an unmistakable general pattern of word-of-mouth recruiting assuredly not aimed specifically at blacks whether or not deliberately aimed to exclude them.
. Of the sixty-five hires, ten were black. Of these, two, or 20%, had listed friends or relatives on their applications. 21 F.E.P. Cases at 1565. Of the ten blacks hired, four were hired as service workers, including the two who listed friends on their applications. Comparing proportions of white and black hires listing friends, and noticing the percentage of black service workers who listed friends, we must eliminate the hypothesis that listing friends was a minor factor: