A total community sample of 3,102 individuals from Evans County, Georgia, was followed for 12-14 years. During this period, 129 documented new cases of cancer were ascertained from medical records and death certificates. Cases were considered for inclusion only if documented at least 12 months after subjects were inducted into the cohort study. Cases were classified as definite, probable, and possible by strict criteria. Blood samples were drawn at the beginning of the study in 1960-62 and sera were frozen. Serum vitamin A (retinol) levels were measured in 1976 on the stored sera of 85 cancer patients and for 174 age-, race-, and sex-matched controls. Retinol estimations were performed by a fluorometric method after alumina column separation. Experiments conducted to simulate the exposure to light, thawing, and refreezing that sera may have undergone during the 14-16 years of storage showed retinol to be quite stable in response to these possible insults. As compared to controls, persons that eventually developed cancer had significantly lower mean serum retinol levels at least 12 months before the cancer diagnosis. The association was in the same direction for all 4 race-sex groups, although stronger overall for males than females, and was consistent for the various cancer sites and cell types. Both matched and regression residual analyses were used to control for the confounding variables considered: age, race, sex, obesity, social class, and smoking.