This study was aimed to validate the 5th and 6th editions of tumor-node-metastasis (TNM) system for patients with hepatocellular carcinoma (HCC), and attempted to improve prognostic stratification by modifying the 6th edition according to vascular invasion and tumor size. From 1986 to 2002, a total of 5,613 HCC cases from Kaohsiung Chang Gung Memorial Hospital in southern Taiwan were enrolled. The 6th edition was modified by dividing stage I into stages IA (single tumor, < or =2cm) and IB (single tumor, >2cm), and by dividing stage II into IIA (multiple tumors, none >5cm) and IIB (tumor with segmental macro vascular invasion). The Akaike information criteria (AIC), within a Cox proportional hazard regression model were used; lower AIC value indicated a better discriminatory ability for staging system. The 1-, 3-, 5-, and 7-year overall survival rates were 45.6, 25.9, 17.9, and 13.4%, respectively. Significant differences in survival curve existed in the 5th, 6th, and modified 6th edition TNM systems. For the modified 6th edition TNM, survival differed significantly between stages IA and IB, and between stage IIA and IIB. The AIC values of 5th (72,328), 6th (72,188), modified 6th (71,991) edition TNM system were decreasing. This investigation demonstrated better prognostic stratifications for the 6th edition than the 5th edition TNM staging system. Moreover, the modified 6th edition staging system demonstrated better prognostic prediction than the former two. Pretreatment staging and simple classification of current modified 6th edition TNM staging can be applied to all HCC patients and are clinically useful.