Medium and long term hydrologic prediction of runoff is one of the most important subjects in the field of water research.This article aimed at predicting runoff at Lancun hydrologic station in the upper reaches of the Fenhe River.Artificial neural network is a nonlinear dynamic system composed of a large number of neurons. Runoff material recorded at the Lancun hydrologic station in the upper reaches of the Fenhe River were used to analyze and predict by three layers back propagation(BP) neural network model. Levenberg-Marquardt method was used to train model. Prediction results meet accuracy requirements, indicating that it is feasible to predict runoff using improved BP neural network model at Lancun hydrologic station in the upper reaches of the Fen River.