Aim There have been no established predictors of the outcome on sorafenib therapy for hepatocellular carcinoma (HCC) patients. We aimed to establish a new prognostic model suitable for sorafenib in HCC. Methods Among 465 HCC patients treated with sorafenib in 14 hospitals, we formed a training cohort with 270 patients at seven hospitals located in West Japan and a validation cohort with 167 patients at seven hospitals located in East Japan. In the training cohort, we examined the relationship between overall survival (OS) and pretreatment clinical factors, and structured a new prognostic model. We verified this model in the validation cohort and compared with four existing staging models. Results Multivariate analysis demonstrated distant metastases, portal invasion, intrahepatic tumor burden of more than 50%, serum α‐fetoprotein of 150 ng/dL or more, des‐γ‐carboxyprothrombin of 1200 mAU/mL or more, albumin of 3.5 g/dL or less and total bilirubin of more than 1.0 mg/dL were significant independent adverse prognostic factors. We calculated a Japan Red Cross (JRC) score with these factors and classified three groups: low‐, intermediate‐ or high‐risk. Their median OS were well stratified (18.0, 8.8 and 3.7 months, respectively, P < 0.001) in the training cohort. In the validation cohort, OS were also statistically stratified (23.9, 10.3 and 2.9 months, P < 0.001). C‐statistics of the JRC score was 0.755, the highest in the five models, indicating its novel predictability. Conclusion Our proposed JRC score well predicts the prognosis of sorafenib therapy, and would be useful to plan individualized strategies for unresectable HCC.