This paper presents the Markov chain methodology to derive the default probability of bank loans. The model is more elaborate than that of the previous model, which was developed by Jarrow, Lando and Turnbull (1997). In our model, through relaxing the assumption of the Jarrow, Lando and Turnbull (1977), we estimate the risk premium only when the default state has not occurred in the stochastic process of non-homogeneous Markov. In our model, the default probability for each borrower and its risk premium will be recursively endogenous. We estimate default probabilities of thirty-one banks in Taiwan. on the other hand, we also compare default probabilities of banks whether participate in financial holding companies or not. The empirical result indicates that banks participate din financial holding companies do not have better credit risk management. Consequently, in facing the BaselⅡAccord, we hope that this paper will be helpful for Taiwan’s financial institutions.