Serum cholesterol concentration was studied for its prediction of cancer in 39,268 men and women aged 15-99 years and initially free from cancer. During a median follow-up of 10 years 1381 cancer cases were diagnosed. Serum cholesterol level was inversely associated with cancer incidence among non-smokers. Age-adjusted relative risks of cancer in quintiles of serum cholesterol were in male non-smokers 1.0, 0.81, 0.73, 0.69, and 0.46 and in female non-smokers 1.0, 0.75, 0.84, 0.78, and 0.70. The associations were not found to be confounded by serum vitamins A or E, serum selenium or several other factors. The association between serum cholesterol level and risk of cancer varied from strongly negative to slightly positive according to subpopulation and site of cancer. The strongest negative associations were found to appear during the first years of follow-up, especially for rapidly developing cancers. Thus the increased occurrence of cancer at low cholesterol levels seems mainly to be due to preclinical cancer.