In financial forecasting system, because there exists a number of factors to effect economy developing, it is difficult to do effective forecasting by using traditional methods such as time series approach, regression analysis. In this paper, a nonlinear model based on RBF neural network is presented. There are some ameliorated measures in learning algorithm of RBF neural network. The number and the centric value of hidden layer are determined by using immune algorithm. The supervisory algorithm is took as method of adjustable weight of output layer. Using above measures, the network is optimized, and the forecast model obtains the precise and objective solution