This paper analyzes high-frequency stochastic volatility (SV) models for the Japanese stock index. Apart from the standard daily SV models, the high-frequency SV models are fit to intraday returns by extensively incorporating intraday volatility patterns. The models consist of the persistent autoregressive stochastic volatility process, intraday seasonality, news announcement effects and correlated jumps in price and volatility. An efficient Bayesian method using MCMC method is developed for the analysis of this model with the multi-move sampler for the autoregressive SV process. Empirical analysis is conducted using the 5-minute returns of Nikkei225 index. Model comparison using BIC indicates that the high-frequency SV models fit the 5-min returns better than the high-frequency GARCH models. They are also shown to perform better in one-day ahead volatility forecasting than the commonly-used realized volatility models such as the heterogeneous autoregressive (HAR) and realized EGARCH models.