Data were analyzed from 33 prospective studies to assess the evidence for a long-term association of low serum cholesterol with cancer. In subjects with cancer diagnosed within two years of the cholesterol measurement or causing death within five years (n = 4,661), the level of serum cholesterol was on average lower than in controls by 0.18 (SE = 0.02) mmol/l in men and 0.11 (SE = 0.04) mmol/l in women; this effect can be attributed to preclinical cancer. For cancers presenting after these intervals (n = 22,030), the average differences were smaller but statistically significant (0.04 [SE = 0.01] mmol/1 [P less than 0.001) in men, and 0.03 [SE = 0.01] mmol/1 [P = 0.005] in women), equivalent to about a 15 percent increase in cancer incidence in the lowest cholesterol quintile. This cannot be attributed entirely to preclinical cancer. In men, there was significant (P = 0.01) heterogeneity between studies as to the extent of a long-term association. The heterogeneity could be substantially explained by socioeconomic status, the association being pronounced in studies of manual workers but absent in studies of professional men. The overall long-term association was attributable mainly to lung cancer in men, and partly to hemopoietic cancers (representing prolongation of survival by treatment). Colon cancer and other cancers unrelated to smoking showed no long-term association with low cholesterol. The data collectively do not justify concern that lowering serum cholesterol to reduce ischemic heart-disease risk might cause cancer. The long-term association with lung cancer is probably caused by smoking and we propose a mechanism.