Mortalities from selected sites of cancer among 5,130 male Japanese physicians followed up for 19 years were examined in relation to smoking and drinking habits surveyed in 1965. With smoking habit classified into three categories (never/past, 1-19 and greater than or equal to 20 cigarettes/day) and drinking habit into four (never/past, occasional, less than 2 and greater than or equal to 2 go of sake/day), the effects of the two factors and their combined effect were analyzed by using the Cox proportional hazard model. Interaction of smoking and drinking was found to be negligible for the sites of cancer studied (upper aerodigestive tract, esophagus, stomach, large bowel, liver and lung), and independent relationships of smoking and drinking with upper aerodigestive cancer were confirmed. Smoking, besides being strongly associated with lung cancer, was also weakly, but significantly, related to stomach cancer. Liver cancer showed no association with smoking whereas this cancer was significantly related to alcohol consumption. The present findings provide further evidence for the association between cigarette smoking and stomach cancer but do not support the relationship recently suggested between smoking and liver cancer.