To determine the extent to which autopsies yield unexpected findings that are relevant to the patient's death and whether cases with a high yield of such findings can be identified selectively, we studied a total of 233 autopsies at a university hospital and at a community hospital. The rates at which autopsies detected major unexpected findings whose premortem diagnosis would probably have improved survival were 11 percent at the university hospital and 12 percent at the community hospital. Major unexpected findings whose premortem diagnosis would not have prolonged survival were found in another 12 and 21 percent of cases, respectively. Pulmonary embolism and fungal infections in immuno-compromised hosts were the most common major unexpected findings. Neither we nor the patients' physicians were able to identify from the clinical data the autopsies likely to have high yields. Furthermore, the physicians' estimates of an autopsy's expected yield were similar for patients evaluated by autopsy and for matched patients who were not. We conclude that the autopsy continues to yield clinically relevant findings at a high level and that it is not currently possible to predict which cases will have high yields. Autopsies are vital to ensure the quality of medical care, and autopsy rates must be increased substantially if this role is to be fully realized.