Abstract: A new method for evaluating commercial bank's internal credit risk based on ARMA-TGARCH-EVT model is proposed. Firstly, a mixed model of ARMA-TGARCH is estimated using GMM, and then the residual series z_t with properties of approximately independent and identically distribution is obtained. Secondly, the POT model of extreme value theory is employed for fitting analysis of the residual sequence to get the estimated value of VAR and ES, and the Bootstrap method is used to determine the confidence interval of VAR and ES at 95% confidence level. Finally, the data that the daily credit asset's logarithmic yields of a commercial bank from 2000-02-19~2010-12-15 are utilized to simulate the results from using this method. Simulation results indicate that, compared with the unadjusted predictive value, the proposed method could overcome the estimation error to some extent, since the sequence's non-independent-and-identically distribution could not meet the assumptions of extreme value theory. Moreover, the method could improve the deviation caused by small samples of extreme events when the likelihood ratio method is used to estimate confidence intervals.