Medication histories written in medical records are important sources of information for clinicians and researchers. Medication histories in outpatient records are frequently inaccurate; the accuracy of the histories in hospital records has not previously been studied. We evaluated the accuracy of medication histories recorded in hospital medical records of 122 persons over the age of 65 at three sites, comparing the written record to a structured history obtained from the patient by the research staff. We defined an error as either the failure to record the use of a medication the patient claimed to use or the recording of a medication that the patient denied using. We found that 83% of all patients had at least one such error, and 46% had three or more. We reanalyzed the data excluding over-the-counter, tropical, and cold medications, and found that 60% of all patients still had at least one error, and 18% had three or more. Failure to record use in the record was more common than recording medications the patient denied using. There were no qualitative differences between hospitals or between the histories obtained by interns and attending physicians. This analysis suggests that medication histories in the hospital medical record are not accurate sources of information in elderly persons. Errors in the history may adversely affect clinical care; researchers relying on hospital medical records to determine medication use at the time of admission should first validate their data.