A series of 1000 cases was selected, on the basis of a clinical and/or post-mortem diagnosis of cancer, out of 4927 autopsies performed at the Institute of Pathologic Anatomy and Histopathology of Turin University. The comparison between clinical and post-mortem diagnoses pointed to an overall concordance with regard to the correct identification of a malignancy as the underlying cause of death of 75%; if the correct identification of type and primary site of the tumor was also taken into account, the concordance was only 56%. The rate of false-positive and false-negative diagnoses, the confirmation rate and sensitivity index of clinical diagnoses, and the error of estimate of the overall frequency of the different types of tumors were computed. Pancreas, liver and biliary tract tumors appear to be the most difficult to identify correctly during life; also lung, stomach and colorectal cancers, lymphomas and leukemias show fairly high rate of clinical errors. Breast cancer, tumors of the nervous system and colorectal cancers appear to be overnotified. These results seem to underscore the necessity of being very careful in drawing conclusions on the frequency and distribution of the different types of cancer on the basis of current mortality statistics.