A multivariate analysis of data from 90 patients with unresectable hepatocellular carcinoma was performed using Cox's regression model to identify factors possibly affecting their prognoses. Thirty-one patients underwent arterial anticancer chemotherapy, and the remaining 59 patients received transcatheter arterial embolization with anticancer agents. Four of 27 variables tested for all the patients (i.e., encapsulation [p less than 0.05], gross appearance of hepatocellular carcinoma [p less than 0.01], clinical stage [p less than 0.01] and therapy [p less than 0.01]) were found to be prognostically significant. Five of 27 variables tested were prognostically significant for the transcatheter arterial embolization group; they were an extension rate of hepatocellular carcinoma (p less than 0.01), encapsulation (p less than 0.01), alpha-fetoprotein (p less than 0.01), prothrombin time (p less than 0.01) and serum sodium (p less than 0.01). Regression equations were used to describe a prognostic index. A prognostic index was defined as the regression equation derived from the results of a total of 90 patients; PI-1 = eY, where PI-1 = prognostic index 1 Y = 1.549 (gross appearance of hepatocellular carcinoma - 1.344) + 0.778 (encapsulation - 1.622) + 0.818 (clinical stage - 1.800) + 1.760 (therapy - 1.344) and prognostic index 2, the regression equation derived from the results of the transcatheter arterial embolization group of patients; PI-2 = eY, where PI-2 = prognostic index 2 Y = 1.210 (extension rate of hepatocellular carcinoma - 1.576) + 1.179 (encapsulation - 1.475) + 0.0001277 (alpha-fetoprotein - 1420.792) -0.039 (prothrombin time - 72.237) - 0.214 (serum sodium - 138.427).(ABSTRACT TRUNCATED AT 250 WORDS)