A methodological study was performed in 1992 to evaluate the accuracy of self-reported use of nonsteroidal antiinflammatory drugs (NSAIDs) and noncontraceptive estrogens that had been dispensed during the previous 12 years. A sample of 560 individuals dispensed NSAIDs or estrogens, and 140 individuals without NSAID/estrogen dispensations were selected from the Group Health Cooperative pharmacy database. Demographic, behavioral, and drug information was ascertained by telephone interview for 356 persons with and 98 persons without NSAID/estrogen dispensations. Of those with only a single NSAID dispensation, 41% (95% confidence interval (CI) 32-50%) were able to recall any NSAID use compared with 85% (95% CI 76-94%) for those with multiple NSAID dispensations. Thirty percent (95% CI 24-36%) recalled the NSAID name, and 15% (95% CI 10-20%) recalled both the name and dose. For estrogens, 78% (95% CI 70-86%) recalled the name, but only 26% (95% CI 17-34%) recalled the name and dose. Age, but not sex, appeared to influence recall accuracy: Persons 50-65 years of age recalled the NSAID name more accurately than those aged 66-80 (odds ratio (OR) = 1.8, 95% confidence interval (CI) 1.0-3.4). A similar advantage was noted for 50- to 65-year-old women in recalling the estrogen name (OR = 1.5, 95% CI 0.6-3.9). Drug name was recalled more frequently for exposures stopped 2-3 years prior to interview than for those stopped 7-11 years prior (OR = 3.0, 95% CI 1.6-5.7, and OR = 2.4, 95% CI 0.9-6.7, for NSAIDs and estrogens, respectively). Specificity was consistently high, ranging from 92% to 100%. This study suggests significant underascertainment of self-reported prescription drug exposure but little evidence that exposures are overreported.