Abstract: Watershed models are cost-effective and powerful tools for evaluating and controlling non-point source pollution, while there is great uncertainty depending on single model simulation which makes multi-model ensemble simulation become inevitable trend. In this study, three nonpoint source pollution models with different structures, named HSPF, SWAT and GWLF, were selected to conduct watershed monthly non-point pollution source ensemble simulation based on Bayesian model averaging (BMA) method, and the expectation maximum algorithm was used to estimate the parameters of the BMA in order to obtain BMA simulation and 90% uncertainty interval. The result using the observed data of the Yixunhe watershed revealed that the Coefficient of Determination and Nash-Sutcliffe Coefficient of the BMA model simulate monthly ammonia nitrogen load and nitrate nitrogen load were higher than that of the single model. The higher the efficiency of a single model is, the greater the weight during the BMA ensemble simulation is. The 90% confidence interval of BMA has a high coverage of measured values in this study. This indicates that the BMA method can not only provide simulation with higher precision through ensemble simulation, but also provide quantitative evaluation of the model structure through confidence interval, which could offer rich information of the non-point source pollution simulation.