To assess the reliability of information reported by patients with AIDS or HIV infection, 123 homosexual men were recruited in Victoria, Australia, in 1992-1993 and interviewed to collect data retrospectively and prospectively on HIV-related illness and treatment. These data were compared with information in medical and pharmacy records. There was no evidence of bias in the patient report of date of diagnosis of HIV infection or date or result of most recent CD4 count, although there was considerable imprecision. Past and current prescription of AZT were recalled accurately by patients compared with medical records (kappa = 0.89 and 0.86, respectively), but previous HIV-related illnesses and enrollment in clinical studies or trials were generally recalled poorly (kappa < 0.4 in some cases). Data collected prospectively on prescription of drugs and use of health care services were reasonably accurately reported by patients. This analysis supports the need for routine assessment of the validity of patient-reported exposure and outcome data in epidemiological studies in order to minimize the possible biases that can result from inaccurate information.