An inverse association between serum total cholesterol and cancer mortality cast a controversy for cause or result of low cholesterol on cancer risk. Therefore, we examined a total of 33,368 Japanese men and women aged 40-69 years, who were free of prior diagnosis of cancer and cardiovascular disease, undertook serum total cholesterol measurement and completed a food frequency questionnaire between 1990 and 1994. They were followed to ascertain incident total and major sites of cancer until the end of 2004 to examine sex-specific associations between cholesterol and cancer risk by incident time, stage and virus infection. After 412,714 person-years of follow-up, 2,728 incident cancers were documented. Serum total cholesterol levels were inversely associated with risk of total cancer in men, with strong inverse associations with stomach cancer in men and liver cancer in both sexes. After exclusion for first 3-year incident cases and advanced cases with metastasis, the inverse association diminished for total and stomach cancers but remained for liver cancer. The multivariable hazard ratios (95% CI) for serum total cholesterol <4.14 mmol/l versus 4.65-5.16 mmol/l were 1.15(0.92-1.43); p-trend across the overall cholesterol categories = 0.25 for total cancer and 1.18(0.79-1.75), p-trend = 0.04 for stomach cancer and 5.12(1.65-15.9), p-trend = 0.0011 for liver cancer in men, and 5.73(1.57-20.9), p-trend = 0.0007 for liver cancer in women. The sustained excess risk of liver cancer associated with low cholesterol was observed regardless of hepatitis-C-virus infection and drinking habits. Although the inverse association for liver cancer remained to be examined further, our findings do not support that low serum total cholesterol levels increase risks of total cancer and other major sites.