A cohort of 8436 men in Taiwan was recruited with personal interview and blood sample collection between 1984 and 1986. During the 5-year follow-up period, 50 incident cases of hepatocellular carcinoma (HCC) were identified. Retinol levels were measured for 35 HCC patients whose serum samples were available and 140 matched controls randomly selected from cohort members without HCC. Lower vegetable intake was significantly associated with an increased risk of HCC after adjustment for other HCC risk factors (P = 0.006). The effect of low vegetable intake on HCC risk was limited to hepatitis B virus chronic carriers and cigarette smokers. As compared with subjects who had a weekly vegetable consumption frequency of six or more meals, the multivariate-adjusted relative risk of HCC for subjects who had a frequency of less than six meals was 4.7 (95% confidence interval, 2.0-11.1; P = 0.0004) among chronic hepatitis B virus carriers and 3.8 (95% confidence interval, 1.7-8.5; P = 0.001) among cigarette smokers. There was an inverse dose-response relationship between the prediagnostic serum retinol level and the development of HCC (trend test, P = 0.003). The odds ratio of HCC for men with a retinol level in the lowest tertile was 9.0 (95% confidence interval, 2.1-39.1) compared with those with a level in the highest tertile. The relation remained after multivariate adjustment for cigarette smoking, habitual alcohol drinking, and either the seropositivity of hepatitis B virus surface antigen and/or anti-hepatitis C virus antibody or the past history of liver diseases through conditional logistic regression analysis. The association was more striking for men 55 years or younger and for those who smoked 10 or more cigarettes/day. There was a significant synergistic effect of hepatitis B virus surface antigen carrier status and low serum retinol level on the development of HCC. These data suggest a potential role of retinol in the chemoprevention of HCC.