Abstract: Based on the approximation theory, this paper presents a new neural network model based on Chebyshev orthogonal multinomial. By theoretical analysis and simulation experiments we show that the neural network has excellent performance of approximating nonlinear systems and the learning speed is high than BP networks.