This paper compares the models in Predicting the VaR of stock price indexes in China. Our results indicate that the normal model usually underestimate the VaR when given probability is 0 01 or 0 02,and the weighted normal model usually overestimate the VaR when given probability is 0 04 or 0 05. The historical simulating model and Logistic distribution model are superior to normal model and to weighted nomal model in predicting the VaR.