Clinical course in hepatocellular carcinoma may be very different. We prospectively evaluated 96 patients with hepatocellular carcinoma unsuitable for radical therapy to investigate factors that could influence survival. Clinical, pathologic, and molecular data of patients were analyzed by univariate and multivariate analysis. The overall actuarial probability of survival at year 1, 2, 3, 4, 5, and 6 was 72%, 41%, 38%, 24%, 20%, and 9%. At univariate analysis, alpha-fetoprotein (AFP) (P =.0082); alkaline phosphatase (P =.0281); bilirubin (P =.0076); etiology (P =.0001); increment of tumor mass at month 3 (P =.0051); type of estrogen receptor (ER) in the tumor (P =.0000); prothrombin time (P =.0003); and portal vein thrombosis (P =.0000) had prognostic significance. At multivariate analysis, only type of ER (P =.0000) and bilirubin (P =.0030) showed independent predictive value for mortality. Survival was significantly longer in patients with wild-type estrogen receptors (P =.0000). Cumulative probability of survival at year 1, 2, 3, 4, 5, and 6 was 94%, 66%, 52%, 43%, 35%, and 18% for wild-type and 51%, 21%, 16%, and 9% for variant estrogen receptors (no patients alive after 4 years). Hepatitis B surface antigen (HBsAg)-positive patients with variant ERs had a median survival of 8 months versus 45 months in anti-hepatitis C virus-positive patients with wild-type ERs (P =.0001). In conclusion, (1) the presence of variant liver ER transcripts in the tumor was the strongest negative predictor of survival in inoperable hepatocellular carcinoma; (2) their presence was associated with spontaneous survival significantly worse than in patients with wild-type estrogen receptors; and (3) HBsAg-positive patients with variant receptors were characterized by the worst survival.