This study examined the validity of self-reported work histories for use in epidemiological studies of toxic exposures. Two uses that were examined were the assignment of subjects to exposure groups and the estimation of exposure duration. Questionnaire responses and data extracted from employment files were compared for 161 retired automobile workers. The questionnaire method assigned subjects to two extreme groups with 99% accuracy but rejected 30% of the subjects because of ambiguous data. Self-reports of exposure duration correlated moderately with the file estimates (r = .63, P = .01). The discrepancy between subjects' estimates and actual exposure duration correlated with subjects' job mobility. The findings indicate that the validity of self-report data declines with the precision required of the data, but that self-reports may have sufficient validity for studies that use extreme-group designs.