Abstract: Resource sharing is one of the key issues of distributed computing, and load balancing is the fundamental approach to sharing scarce resource in a distributed computing system. However, the existing load balancing methods are mostly confined to the homogeneous networks. With the diversity of the computing terminals, there is increasing requirement for the study of load balancing toward the heterogeneous networks. In this paper, a diffusion-based dynamic load balancing algorithm is proposed for the heterogeneous networks and it is proved with mathematical strictness that all nodes will converge to the excepted balanced point. The numerical results show that the algorithm is better than the GDA algorithm proposed by Rotaru et al. (2004) and has ideal convergence property over many network topologies including Mesh, Star and Tours etc. And it converges rather fast even in a randomly generated network.