AIM: To study the practical use of the model for endstage liver disease (MELD) on clinic and assess its validity by the concordance (C)-statistic in predicting the prognosis of the patient with severe viral hepatitis.METHODS: One hundred and twenty-one patients were divided into plasma exchange group and non-plasma exchange group, and were graded with MELD formula.The death rate was observed within 3 mo.RESULTS: Eighty-one patients died within 3 mo (35 cases in PE group, 46 cases in non-PE group). The mortality of patients in PE group whose MELD score between 20-30and 30-40 were 31.6% and 57.7%, respectively, but in non-PE cases they were 67.6%, 81.3% respectively.There was significant difference between PE group and non-PE group (P＜0.05). However, the mortality of patients whose MELD score higher than 40 were 93.3% in PE group and 100% in non-PE group and there was no significant difference between the two groups (P= 0.65＞0.05). The optimal cut-off values of MELD to predict the prognosis of patients were 30 in PE group whose sensitivity, specificity and C-statistic were 80.0%, 52.0% and 0.777, but in non-PE group they were 25, 82.6%, 86.7% and 0.869, respectively.CONCLUSION: The MELD score can act as a disease severity index for patients with severe viral hepatitis, and the mortality of the patient increases with the increase of the MELD score. The MELD can accurately predict the short-term prognosis of patients with severe viral hepatitis.