Abstract: Neural networks show good ability of self-adaption and self-learning, and is widely applied in non-linear black-box modeling and system identification. This paper proposes a wavelet neural networks based system identification method. Because of the time frenqency location characteristic and multi-scale ability of wavelet transformation and muli-resolution analysis, wavelet is used as the basic of the basis neural networks, and we call it wavelet neural networks. First the paper gives the general framework of the proplem of non-linear black-box modeling system identificaiton, then uses wavelet neural networks as the black-box model, and at last the computer simulation results show that the method is practical.