In order to further improve the accuracy of remote sensing retrieval of inland water quality, the paper chose Wenyu River in Beijing as research object,and used ETM+ data and plesiochronous measured water quality parameters(turbidity, BOD5) data to establish BP neural network models with several hidden layers being one. It chose the best neural network model aimed at turbidity and BOD5 and used ETM+image. to retrieve turbidity values and BOD5 concentration values. Finally, it compared the retrieval results with the result retrieved by the conventional multiple linear regression model.The result shows that the remote sensing retrieval of Wenyu River water quality is a nonlinear problem, using BP neural network method for both turbidity and BOD5 water quality retrieval is superior to the linear regression method.