This study involved 97 male veterans (age range, 45-65) in four approximately equal groups, classified as follows: 1) colon or rectal cancer, 2) upper digestive-tract cancers, 3) other cancers, and 4) additional controls (no cancer). Dietary histories of these cancer and noncancer inpatients were recorded to see whether or not consumption of particular substances could be predictive of cancer. The interactions of substances in combination such as sugar x fat, and alcohol x smoking, proved to be important predictors of cancer of the upper digestive tract. The degree of coffee consumption was a predictor of nondigestive-tract cancer. However, colorectal cancer patients could not be distinguished from non-cancer patients by the dietary variables examined.