Analysis of the clinical and post-mortem assessment of the underlying cause of death in 742 autopsies showed that over- and underdiagnosis cancelled each other out in the majority of the main diagnostic groups, so there was little difference in the total number of cases recorded in the different groups after clinical and post-mortem investigation. However, in the individual case the reliability of the clinical diagnosis varied greatly with the nature of the diagnosis and its degree of certainty. Reliability was for example high with clinically certain arteriosclerotic heart disease and low with cerebrovascular disease. Underdiagnosis of lung cancer is still a problem. Lack of interest in autopsy investigation may be reflection of lack of clinical involvement or therapeutic frustration rather than the use of sophisticated diagnostic procedures, as has been assumed by previous authors. It is suggested that the idea of selection of cases for autopsy should be replaced by selection of autopsies for microscopic investigation on the basis of the macroscopic post-mortem findings. It is suggested also that clinicians might profitably attend autopsies on patients in their sphere of interest that were not admitted under their care.