Load balancing is a resource managing block in a distributed system, whose main function is distributing system load logically and pellucidly between servers in order to achieve the optimization of the integrated performances. Load balancing technology can be integrated into middleware, which can be used in heterogeneous systems with access transparency and extensible capability. In the middle module we can achieve a variety of load balancing algorithms flexibly, in order to meet different load balancing demands of applications, and make the systems become more scalable, configurable, and robust. This paper designs a load balancer based on Web Services, and actualizes a dynamical load balancing scheduling policy according to the model of distributed systems. In order to describe system load more accurately, this paper chooses the CPU utilization, memory utilization, system response time, input and output streams and the process number as the parameters to measure the system load. In order to avoid increasing pressure on the balancer, this strategy will make the load information collection program run on the server nodes. When the value of the load difference from two interfacing acquisition cycles exceeds the standard, the collection program will call the Web Services on the balancer, reporting its own load status. And the balancer runs the synchronized operation according to the information collected from all the loads, and replaces the service sequence. Finally, the simulation experiment results shows that the dynamical load balancing scheduling policy performs well.