In a program screening civilian applicants for U.S. military service for human immunodeficiency virus (HIV) infection, we studied the frequency of false positive diagnoses retrospectively among applicants seropositive for HIV in a subpopulation with a very low prevalence of infection. That subpopulation was defined as consisting of all applicants tested between October 16, 1985, and June 30, 1987, who were young (17 or 18 years of age) and resided in a rural county in a state with a low incidence of reported acquired immunodeficiency syndrome (n = 135,187). Serum specimens from 15 applicants positive for HIV in this low-prevalence subpopulation were retrieved from a serum bank and retested by two Western blot methods, radioimmunoprecipitation, and an immunoassay constructed from a molecularly cloned and expressed viral envelope polypeptide. Fourteen of the 15 samples were unequivocally positive on all retest assays, and 1 was negative. Thus, the measured rate of false positive diagnoses in this program was 1 in 135,187 persons tested. Factors important in achieving a low false positive rate were redundant, multistep testing algorithm, conservative criteria for interpreting Western blots, the requirement that a second, newly drawn serum specimen be tested for verification before a diagnosis of HIV was considered established, and tight quality control of laboratory testing procedures. We conclude that a screening program for HIV infection in a low-prevalence population can have an acceptably low false positive rate.