To master the variation of water quality in case of water security event and to take measures in advance against that in Xiangjiang river basin, the monitoring data of DO and NH4+–N in Changsha section and Yiyang section, which are two serious pollution river sections in the basin, were adopted for predicting their contents through ARIMA model which infers a classical time series model using Bayesian approach, the model parameters and prediction results were simulated by employing Markov Chain Monte Carlo (MCMC) method. The results showed that Bayesian approach in the model could accurately predict contents of DO and NH4+–N at section level, interval level, and probability level in the two selected sections.