This paper used Back-propagation(BP) algorithms and Davidon least squares-based learning algorithm to train the neural network(NN) to predict the nonlinear self-similar network traffic respectively.The feasibility and advantage of these two algorithms were discussed by analyzing the Mean learning errors,training errors and the convergent speed of these two training algorithms.The simulation demonstrated that the NN trained by both of these two training algorithms can well predict this traffic.Compared with BP algorithms,the Davidon least squares-based learning algorithm can converge quickly and has the almost same prediction accuracy.It supplied a feasible method to predict the complex self-similar network traffic.