Abstract: In order to provide foundation for making scientific irrigation schedule and heighten the water use efficiency, the advantage of BP network and wavelet analysis is integrated to establish a wavelet BP network model for predicting water requirement of rice. Replaced the traditional S-shaped activation function with the wavelet function, 16 years actual measuring series of water requirement of rice in Fujin irrigation area of Sanjiang Plain were analyzed. With the network structure of 6-10-1, the precision of prediction reaches 0.01 while it is disciplined 355 times. The result is superior to both BP algorithm and RAGASABP model, and shows that the model has fast convergence speed and high forecasting precision. A new idea of research is given in the field.