Nutrients play important roles in human carcinogenesis. To investigate the relationship between serum retinol level and subsequent risk of liver cancer, a nested case-control study was conducted within a cohort of 9775 male adults recruited from 1984 to 1986. A total of 35 liver cases newly diagnosed during the follow-up period were identified, and 140 cancer-free controls matched to liver cancer cases on age, residential area, and time of blood collection were also randomly selected. The blood samples collected in 1984-1986 from all study subjects were stored at -30℃ until examinations. Serum retinol levels of cases and controls were determined by high performance liquid chromatography blindly. There was a significant association between serum retinol level and risk of liver cancer. After adjustment for HBsAg carrier status, darkgreen and yollow vegotable consumption frequency, vegetarian habit; the odds ratio of developing liver cancer for those who had the highest serum retinol level was 0.03 (95% confidence interval = 0.002~0.42) compared with those who had the lowest level as the referent. There was also a statistically significant dose-response relation between serum retinol level and liver cancer risk. In the further analysis of data for cases who developed liver cancer after 1987 and their matched controls, the association remained unchanged. After adjustment for HBsAg carrier status, serum retinol level, and vegetarian habit; the odds ratio of developing liver cancer for those who consumed darkgreen and yollow vegotable less than ten meals per week was 9.95 (95% confidence interval=1.42~69.9) compared with those who consumed more than ten meals a week as the referent.