Comparison of certified clinical diagnoses with autopsy findings showed that, while the major cause of death was confirmed in 61 per cent. of cases, many diagnoses--both major and contributory--were wrong; many clinical diagnoses were either disproved or relegated to a less important role, and many autopsy findings had not apparently been anticipated. Accuracy was particularly poor in some clinical categories: notably cerebrovascular disease and infections. In these, the diagnosis was more often wrong than right. Thus, death certificates are unreliable as a source of diagnostic data. The clinician's confidence in his major diagnosis bore a fairly close relationship to the frequency of its confirmation. Nevertheless, even when certified as "fairly certain", the major diagnosis was wrong in about one-quarter of these cases. An attempt was made to assess the significance of incorrect diagnoses; one half of these might be clinically significant. Diagnostic accuracy did not improve with the time spent in hospital, and it bore an inverse relationship to the patient's age.