Abstract: A model of Artificial Neural Network with Genic Arithmetic(GA-ANN) is established to predict reference evapotranspiration.This model integrates the merits of seeking for a global optimum solution by using genic arithmetic and the well-mapping capacity of the back propagation neural network.It can determinate the optimum model structure automatically.Eight groups of model input factors' composition are set up,and their correlative influence on the model's forecasting precision are studied.An optimum model structure for predicting short time period(daily and decade) reference evapotranspiration is present,in which only daily mean temperature,sunshine hours and the Julian day's ordinal number are considered as the input factors.A case study shows that the model overcomes the disadvantages in determinating the model structure when using back propagation neural network.And it has high precision with good adaptability and feasibility.