To investigate the predictive ability of overnight information on daytime trading in Chinese commodity futures market，this paper gives the stochastic volatility models based on normal，student-t，generalized error and the mixture of normal distributions respectively．Then，using Bayesian Markov Chain Monte Carlo (MCMC) estimation techniques，the empirical analyses are given for copper，aluminum，soybean and wheat futures markets. The results show that the stochastic volatility models with the mixture of normal distributions can better fit the stochastic volatility with normal，student-t，generalized error distributions in describing the impact of overnight information on daytime trading prices．Total overnight information shows significant predictive ability for daytime returns and volatility．Moreover，the aspects and degrees of prediction are different for different futures contracts．Furthermore，weeknight returns，weekend returns and medium-long holiday returns show prominent predictive abilities for daytime returns and volatility．Their predictive abilities are evidently stronger than those of the total overnight information，and show different degrees of asymmetry Concretely，there are leverage effects for the impact of overnight information during the weeknights，weekends and medium-long holidays on the daytime returns and volatility in most futures contracts，except for inverse leverage effects of medium-long holiday in both soybean and wheat futures contracts．