Facing high investment risk, investors shall evaluate the performance of their portfolio based on Value at Risk (VaR) model. To set up appropriate VaR model, the researchers must examine data characteristics and window period; but most current studies treat different portfolio with same VaR model and window period. Normal test and back test are implemented in this study to determine the optimal model and window period of different types of mutual funds. Four kinds of mutual funds (domestic, foreign, bond and balanced) and three level of performance (good, normal and bad) are considered. Contrasting to previous studies, this study can offer more accurate VaR model and window perod for investment decisions. The main findings are (1) the return distributions of all the 48 sampled mutual funds do not fit normal distribution, (2) each type of mutual fund has a different optimal model and window period and bond fund needs the longest window period, (3) the rank of performance evaluation of mutual fund is different when based on either raw return or VaR, (4) the performance persistence of different type of mutual fund is not the same; domestic equity fund and foreign equity fund show adverse change, bond fund and balanced fund show persistent performance.