In a university Department of Internal Medicine, we compared the clinical information, as written down immediately post mortem, with the demand for autopsy and with the autopsy findings during one complete year. Autopsies were performed in 143 of 306 consecutive deaths. In autopsied patients clinicians had recorded more diagnoses per patient, but with less certainty than in non-autopsied patients. Patients with a diagnosis of a solid tumour were autopsied less often, and patients with infections or gastrointestinal disease more often than the average patient. These findings indicate selection of cases for autopsy. In 41% of patients erroneous diagnoses were detected that might have influenced management of the patient. In 28% of patients autopsy offered no additional information about one or more diagnoses. Both the proportion of errors and the proportion of uninformative autopsies increased with decreasing clinical certainty about the diagnosis. We conclude that autopsy offers useful information in most cases, and that selection of cases for autopsy is probably not justified.