The Japanese staging system that is generally used for hepatocellular carcinoma (HCC) (3rd edition) was considerably revised recently, especially T category. No study, however, has revealed how well the new classification (4th edition) works to stratify HCC patients at a pre-intervention stage. The purpose of this study is to assess the discriminatory value and predictive power of the 4th edition, and to compare its utility with the clinical utilities of the 3rd edition and the cancer of the liver Italian program (CLIP) score, as determined from 662 Japanese patients. We performed a retrospective analysis of the HCC diagnoses at four Japanese institutions from 1990 and 1998. Overall survival was the only end-point used in the analysis. The discriminatory ability and homogeneity of the 4th edition were compared to those of the 3rd edition and the CLIP score. As of January 1999, 440 patients (66.4%) had died. The overall median survival was 37.7 months. Liver function and tumor variables, which are already accounted for by the different scoring systems, were significantly associated with survival. Compared with the 3rd edition, the 4th editions' discriminatory ability (tested by the linear trend test) and homogeneity of survival within each category (tested by the likelihood ratio test) were enhanced in both the overall group of patients and the subgroups of patients receiving transcatheter arterial chemoembolizations and percutaneous ethanol injections. In patients receiving surgery, however, the 4th edition's abilities were the lowest among the indices. Multivariate analysis revealed that the CLIP score that includes liver function had additional explanatory power above that of the 4th edition. These findings indicate that the 4th edition has a higher stratification value than the 3rd edition. However, this benefit is due to the non-surgical patients, rather than to the surgical patients. If the 4th edition had an additional scoring system based on its original tumor staging and liver damage, it might be highly beneficial, although relative risk ratios of those should be analyzed.