At present, the extendible Web server clusters are the important framers for the high performance network servers. The architecture based on load balancer is often used on web server clusters. The load balancer is the main bottleneck in improving performance of cluster. For it is hard to complete the task of assigning the requests from the client while collecting the real load information of the web servers efficiently. In this paper, we propose a model of two levels load balancer for web cluster to resolve the contradiction. When web requests are arrive, the first-level balancer continuously assigned them by to the second-level which consist of some load balancers connect to real web servers and receive real-time load information of them. Round robin algorithm is used at the first-level load balancer to schedule the client’s requests in order to shorten the response time. The proposed algorithm in our model is run at the second-level load balancers to schedule the requests to real web servers according to the real load information. At last, a simple test is carried out to make sure the performance of the model. The result of the test shows that the model has made better load balancing.