Distributed applications use predictions of network traffic to sustain their performance by adapting their behaviors. It has been recognized that the network traffic consists of a majority of linear part and a small quantity of non-linear part which can not be neglected. However, existent network traffic prediction algorithms only utilize either linear or non-linear methods to solve the problem and can not provide enough accuracy and realtime due to the isolated adoption. A hybrid network traffic prediction algorithm, is provided, in which Kalman filter (KF) and wavelet are combined. Thus the linear part can be processed by KF and the non-linear part can be done by wavelet. Simulation results show that the proposed algorithm can guarantee higher accuracy and better realtime than those algorithms based singly on linear or non-linear method.