This paper proposes a data mining classification algorithm named SRmix (the mixture of SVM and RBF network) which can enhance the generalization ability of classical RBF network. The model is based on the idea of complementarity and cooperation. The novelty lies in using support vector machine (SVM) to determine the relevant key parameters of Radial Basis Function Neural Network (RBFNN) and elaborating the relationship between these two models in different aspects. The experimental results also show that SRmix performs satisfactory in algorithm stability and accuracy.