The purpose of this article is to explore, by Bayes Classifier, how the insurance payments perform in the light of the reciprocal attributes of the people involved in the given situation. At first, we use the k2 algorithm to train the Bayes Network model, and compare the results with those of the traditional taxonomy (such as cart, logistic) and the Naive Bayes model. Because of the limitations of k2, however, we put forward and adopt the Bayes Network model on the basis of associate rule, resulting in a model with better robustness, which also guarantees the capability of forecasting.