Taking highfrequency data of CSI300 index as example, the outofsample volatility predictions and a SPA test are used to evaluate the predicting ability for different historical volatility models and realized volatility models.The empirical results show that realized volatility model and the extended SV model are superior to other models. However the GARCH model and its extended type, which is popular in financial academe and practice, perform the worst for volatility predicting of Chinese A-share market.