Records from 910 autopsies performed at a university hospital in Salvador, Bahia, Brazil, were examined in order to assess the accuracy of clinical diagnoses of the patients' underlying causes of death. This study found inaccurate clinical diagnoses in 31% of the cases. The overall rate of diagnostic error appeared to remain fairly stable from 1970 to 1982, being highest for older patients. Thirty-six percent of the 263 cancer deaths were incorrectly diagnosed, and a number of pathologies considered relatively easy to diagnose were not always correctly identified. Quite aside from their direct medical implications, diagnostic errors of the magnitude observed in this and other studies seriously jeopardize the quality of vital statistics and such statistics' usefulness for improving public health.