The Recursive Orthogonal Least Squares (ROLS) algorithm is applied to train Radial Basis Function Neural Network (RBFNN) when modeling, so as to save large memory and computational efforts. Using the information available from the trained network with ROLS algorithm, the effective centers of network can be obtained by adopting backward selection algorithm, which achieve acceptable accuracy with significant reduction of network structure. The results of simulation and experimentation show that the algorithm is efficient and useful.