Summary. The objective of this study was to determine the predictive value of the model for end-stage liver disease (MELD) scoring system in patients with acute-on-chronic hepatitis B liver failure (ACLF-HBV), and to establish a new model for predicting the prognosis of ACLF-HBV. A total of 204 adult patients with ACLF-HBV were retrospectively recruited between July 1, 2002 and December 31, 2004. The MELD scores were calculated according to the widely accepted formula. The 3-month mortality was calculated. The validity of the MELD model was determined by means of the concordance (c) statistic. Clinical data and biochemical values were included in the multivariate logistic regression analysis based on which the regression model for predicting prognosis was established. The receiver-operating characteristic curves were drawn for the MELD scoring system and the new regression model and the areas under the curves (AUC) were compared by the z-test. The 3-month mortality rate was 57.8%. The mean MELD score for the patients who died was significantly greater than those who survived beyond 3 months (28.7 vs 22.4, P = 0.003). The concordance (c) statistic (equivalent to the AUC) for the MELD scoring system predicting 3-month mortality was 0.709 (SE = 0.036, P < 0.001, 95% confidence interval 0.638–0.780). The independent factors predicting prognosis were hepatorenal syndrome (P < 0.001), liver cirrhosis (P = 0.009), HBeAg (P = 0.013), albumin (P = 0.028) and prothrombin activity (P = 0.011) as identified in multivariate logistic regression analysis. The regression model that was constructed by the logistic regression analysis produced a greater prognostic value (c = 0.891) than the MELD scoring system (z = 4.333, P < 0.001). The MELD scoring system is a promising and useful predictor for 3-month mortality of ACLF-HBV patients. Hepatorenal syndrome, liver cirrhosis, HBeAg, albumin and prothrombin activity are independent factors affecting the 3-month mortality. The newly established logistic regression model appears to be superior to the MELD scoring system in predicting 3-month mortality in patients with ACLF-HBV.