Abstract:This paper introduces a hybrid learning algorithm for Radial Basis Function neural network(RBFNN) based on subtractive clustering, K-means clustering and particle swarm optimization algorithm(PSO). The algorithm can be used to determine the number of hidden layer nodes and initial clustering centers of K-means by using subtractive clustering; Then the initial particle swarm of PSO can be formed by K-means clustering algorithm．The basic PSO algorithm are optimized and developed to improving convergence and stability of the algorithm, and finally the improved PSO algorithm is used to train all the parameters of RBFNN. The simulation for IRIS data set classification problem is executed, the experiment results show that the improved hybrid algorithm has higher accuracy and better stability than several other popular methods.