For the problem that it is difficult to determine the hidden layer structure of the radial basis function(RBF) neural network, based on the good online classified characteristic of adaptive resonance theory(ART) neural network, a self-organizing RBF neural network structure design algorithm is proposed. The algorithm uses the clustering characteristic of ART neural network to design the RBF neural network structure. Through the similarity comparison of input vector, the number of the hidden layer nodes and initial parameters are determined, so that the network has simplified structure. The experiment results show that the proposed structure has a smaller number of nodes, fast learning speed and better approximation ability.