Abstract: The internal rating-based approach is the core content of New Basel Accord. The calculation of probability of default, loss given default, expected losses and other concerning factors are the key steps to bring internal rating-based approach into effect. Based on the practical data of our state-owned commercial banks, a relative scientific evaluating system is established in this paper by stepwise discriminant analysis, and a probability of default forecasting model is constructed by Bayes discriminant model. Also expected losses are calculated by neural network based on Levenberg-Marquardt algorithm. Therefore, loss given default could be work out by the function among probability of default, loss given default and expected losses. Empirical results show that this model could be of certain validity and feasibility to forecast probability of default and loss given default.