To obtain a better capacity of clustered Web servers, a load balancing algorithm is presented in this paper based on dual-load prediction, using the different characteristics of Web server users' requests for static and dynamic content. First, wavelet packet-SVR (support vector regression) prediction model is constructed; then, the arriving assignments are distributed according to the type of the requests and the load predicted. The simulation result proved that the proposed method has better performance then traditional load balancing algorithm based on load prediction.