The present study was performed to investigate the relationship between diabetes mellitus (DM) and primary hepatocellular carcinoma (HCC). Incident HCC cases were recruited in Kyushu, Japan. Ethnicity-, age-, gender-, residence-matched hospital controls and community controls were collected. Information on viral hepatitis B (HBsAg) or viral hepatitis C infection, history of blood transfusion, past histories including DM, amount of drinking or smoking, and genotypes of alcohol metabolizing enzymes was collected. Associations between these items and HCC were analyzed multivariately by conditional logistic regression analysis. Two hundred and twenty two (177 males and 45 females) case-control sets were completed between July 1995 and June 2000. Since hospital controls turned out to be a biased one or those sampled from a DM-prone population, a multivariate analysis was performed for the HCC-community controls sets, and it yielded significantly elevated odds ratio (OR)s due to past histories of DM (2.522; 95% Confidence Interval (CI) = 1.267-5.020), blood transfusion (1.747; 1.136-2.689), and unit increment of alcohol consumption (1.358; 1.096-1.684) for males. The same analyses of the HCC-community-controls sets for females, revealed an elevated but not statistically significant OR due to past histories of DM (4.195; 0.808-21.805). A multivariate analysis revealed that DM might be a risk factor for HCC.