【Objective】The present study was conducted to predict water pollutant concentrations in Taihu Lake using BP neural network model in order to find out the mechanism of changes in water pollutant concentrations in lakes. 【Method】The BP neural network forecast method for predicting water pollutant concentration in Taihu Lake was established on the basis of 4 water quality indices， viz.， pH value， dissolved oxygen， CODMn and NH3-N from 2004-2010. The data were obtained from section automatic monitoring station of Wang River （Jiaxing， Zhejiang）. The water quality of Wang River during first 5 weeks of 2012 was predicted by using the established model. 【Result】In order to simplify the structure of BP neural network model and improve the prediction speed， the study established a three-layer BP neural network model. The predicted results were found accurate and the model was found to efficiently predict the changes in water pollutant concentration in lakes. The predicted results of water quality of Wang River during first 5 weeks of 2012 showed that the water quality of Wang River belonged to Ⅴ grade and became worse. The prediction results were in accordance with the development trend of water pollutants in Taihu Lake. 【Conclusion】BP neural network model had good nonlinear mapping capability and flexible network structure. It could better reflect the changes patterns of water quality index with high prediction precision and provide a scientific prediction mechanism for controlling water pollution.