Physicians who think epidemiologically are rare. A method is suggested for detecting their aptitude early in their career when help may be offered to make the most of their special talent. Clusters geographically or in families may provide clues to cancer etiology. Clusters have been systematically thought by mapping cancer mortality in the US and independently in China. Case-control studies have revealed environmental exposure responsible for some of the clusters. Clusters noted by alert clinicians or other astute observers have revealed most of the known environmental causes of human cancers. Genetic influence in carcinogenesis has been identified by studies of peculiar cancer occurrence, such as familial aggregation, multiple primary cancer or the occurrence of cancer with other diseases as, for example, congenital malformations and immunodeficiency disorders. Ethnic differences in cancer occurrence may be revealing. Thus, in Japan there is low frequency of B-cell lymphoma but high frequency of certain autoimmune diseases, as if inherent protection against one predisposes the other. As a rule of thumb, the occurrence of three rare observations is not likely to be due to chance. Examples include ideal carcinoma in three persons with cystic fibrosis of the pancreas who survived to about 30 years of age, and the occurrence in Klinefelter's syndrome of germ cell tumor of the pineal--a neoplasm that has an unusually high frequency in Japan. Finally, the history of discoveries concerning cancer etiology, an aspect of what Comroe has called "research on research", can point the way to new discoveries in the future.