Web server clusters has become a popular solution to achieve high performance due to their low price, high reliability, and good scalability. One of the key technologies is “request dispatching”, which is to centrally accept all the incoming HTTP requests and dispatch them to the servers in the cluster to achieve parallelism The research shows that the efficiency of request dispatching is determined by whether the server loads are balanced during the dispatching. spatching. This paper sudes the characteristics of the load generated by HTTP requests, gives an efficient method of load sampling, and preserts a novel request dispatching algorithm which can predict the load and balance it. The method can be used to build Web server clusters of heterogeneous machines.