CVaR will be taken into the RAROC model, which can improve the RAROC model in the evaluation of funds' performance. This article uses seven years of data from 2005 to 2012 of open-end funds which is founded before 2005. The calculations of VaR are in two methods. One is Modified Cornish-Fisher in nonparametric estimation, the other is generalized autoregressive conditional heteroskedasticity GARCH model in parameter estimation. The residuals are subjected to T and GED distributions. The results of two methods take a comparison through the return of the test. Through empirical testing, Modified Cornish-Fisher can measure the risk better with the low confidence level and GARCH model subjected to GED distribution can measure the risk better with the high confidence level. Then, calculate the CVaR and the results of VaR and CVaR are compared to test. The conclusions show that CVaR can cover the loss better and the accuracy of RAROC based on CVaR increases, more accurate ranking, which provides a very good indicator of the performance for investors.