We studied the relation of total serum cholesterol to all cancer and site-specific cancer incidence in a cohort based on a probability sample of the United States population. A total of 5125 men (yielding 459 cancers) and 7363 women (398 cancers) were initially examined in 1971-75 and followed a median of 10 yr. An examination of age-adjusted incidence rates by cholesterol level showed an inverse association between cholesterol and all cancer; lung, colorectal, pancreatic, and bladder cancers; and leukemia. In women a weak inverse relation (reflecting an elevated rate among those only in the lowest cholesterol quintile) was apparent for all cancer; more prominent inverse associations were seen for cancers of the lung, pancreas, bladder, cervix, and for leukemia. A more detailed analysis of cholesterol and colorectal cancer revealed little association in both men and women. For an aggregate group of smoking-related cancers, the inverse relation was especially prominent: the multivariate relative risk estimates for subjects in the lowest cholesterol quintile, compared to those in the highest quintile, were 2.1 (1.1-3.8) and 3.3 (1.4-7.8) for men and women, respectively. The inverse association was present for smoking-related cancers diagnosed 6 or more yr after cholesterol determination in both men and women, suggesting that this association cannot be simply dismissed as a preclinical cancer effect. Further investigation of the cholesterol-cancer question, particularly the relation between cholesterol and smoking-related cancers, may provide useful etiological leads.