The protocols of 2,145 autopsies were retrospectively reviewed and the findings compared with the clinical diagnoses. A sudden decline in the autopsy rate that occurred during the period studied was followed by a highly statistically significant difference in clinical accuracy (P less than 0.01), in favor of the predecline period. The overall rate of major discrepancies was 29 per cent. The most frequently missed diagnoses were infections, which were found in 26 per cent of all autopsies and had not been diagnosed clinically in 63 per cent of these cases. Malignancies occupied second place among overlooked diagnoses in the selected disease categories; in 99 per cent of the cases the malignancy was the principal diagnosis, and it had been misdiagnosed clinically in 42 per cent of these cases. Cerebrovascular disorders were correctly diagnosed in most cases (87 per cent of the patients in this group). Among autopsy diagnoses labeled as the immediate causes of death, the most frequently overlooked were pulmonary embolism and gastrointestinal hemorrhage, which were not recognized in 84 and 78 per cent, respectively. In cases in which clinicians were not entirely confident in their impressions, their diagnoses were usually confirmed at autopsy. In these cases 15 per cent of the patients died soon after admission to the hospital, with accurate diagnoses in 71 per cent. The discrepancies disclosed should be regarded as sufficiently large to mandate continued emphasis on autopsy evaluation as the basis for the control of the quality of patient care.