A stepwise regression new algorithm for the structure determination of radial basis function networks is proposed. In this algorithm, based on the orthogonalization method and the "Innovation-Contribution" criterion, thc structure of radial basis function networks and parameter estimation are simultane-ously determined. It not only avoids ill-conditioned problems occured in many common leaming algorithm which is often based on first choosing randomly some data points as radial basis function centers and then using singular value decomposition to solve for the weights of the network，but also can reduces considerably the amount of computations and we can obtain the optimal radial basis function networks．Finally，the identification of a simulated system show the correctness and effectiveness of this new algorithm．