Abstract: Since every processor may be different and sources are shared in distributed memory systems, the time cost of a task on a processor is stochastic. In this paper a new parallel scheduling algorithm on distributed memory systems is brought forward, which mainly introduces a character parameter according to characters of systems and tasks, and adopts the overlapping technology of communication and computing. It not only gets load balanced very well and utilizes sources sufficiently, but also effectively improves the speedup ratio even if in the multi user conditions. Experiments show that the scheduling algorithm put forward in this paper is much more effective to utilize systems sources and raises higher speedup ratio than other similar algorithms.