Abstract: Neural network ensemble is a very popular learning paradigm where the outputs of a set of separately trained neural network are combined to form one unified prediction,and it can significantly improve the generalization ability of the learning systems.To improve the effectiveness of ensemble,a RBF networks ensemble based on the divided and ruled thought and an island migrating model is presented in this paper.Experimental results show that the predictive model of neural networks ensemble based on an island migrating model can not only raise the system’s searching ability of high dimension for the much dimension space,simplify the structure of the networks,but also can gain higher predicting accuracy in the automatic examination．