Abstract: A hybrid learning algorithm for wavelet networks is presented to quicken up the speed of convergence, which combines the Levenberg-Marquardt (LM) algorithm with least squares methods. The LM algorithm trains a wavelet network over a reduced weight space that consists of nonlinear parameters of the wavelet networks. The remained weights that are linear parameters are calculated in accordance with least square methods. Identification of a chaotic system is applied to demonstration of the performance of the proposed hybrid learning algorithm. The results indicate that the proposed algorithm is very efficient and enables the learning process to significantly speed up.