Taking round-robin algorithm to determine the optimal BP neural network structure, the paper established BP neural network model to predict water quality. In view of the shortcomings such as lower learning convergence speed of BP neural network, easy to fall into local extremum, in the same conditions of network structure and expectation error,the paper used GA to optimize the initial weights and threshold of BP neural network, and build GA-BP and multi-hidden layer BP neural network prediction model for water quality.The paper took total nitrogen of a reservoir in Yunnan Province for example to predict , compare and analyze. The results showed that ①prediction accuracy of GA-BP network model is better than that of water quality model of basic BP network, indicating that the genetic algorithm can effectively optimize the BP network initial weights and thresholds. ②The increase of hidden layers BP neural network can further improve the network prediction accuracy, but further extend the training time. ③GA-BP and a number of hidden-layer BP network can improve the prediction accuracy as an effective way, both can be used to forecast water quality, and provide new ways and methods for water quality forecast. In contrast, the faster convergence speed and higher prediction accuracy of GA-BP model have a certain computational advantages.