Abstract: For the multi-scale characteristic of network traffic, a prediction algorithm based on wavelet transform and combinational SARIMA model was introduced. The network history traffic was decomposed with wavelet method to analyze complex correlation structure of the network traffic. According to the periodicity and self-similarity of the traffic series under different time scales, different prediction models were selected for prediction. The resulted series was reconstructed with wavelet method to get the results of original traffic. Simulation results showed that the proposed method can achieve higher prediction accuracy than that of the similar prediction methods.