Whether high-frequency fluctuations in stock price data is helpful to improve the prediction accuracy is the forefront issue of academic. Taking the SSEC index and SP500 index as sample, we compute the volatility predicting results based on volatility models with different sampling frequency and out-of-sample rolling time windows method. Using bootstrapping SPA test, we compare the predicting performance of different volatility models. The empirical results show that, considering the skewed, leptokurtic, and fat tailed distribution in stock market returns, volatility model based on high-frequency return data outperforms models with daily return data when volatility prediction accuracy is concerned.