Static task scheduling on network of workstations is well-known to be an NP-co mplete problem in a strong sense. Some heuristic algorithms have been proven to be sub-optimal under some restrictive conditions. In this paper, the authors pr esent a heuristic algorithm named DAG (directed acyclic graph) partition and sub -graph reconfiguration algorithm, which is a fast and effective one used in par allel task scheduling. The complexity of this algorithm is O(log|V|I1518 ×(|V|+|E|)). It adopts recursion to implement DAG partition and sub-graph re configuration, then builds task clusters to carry out the task scheduling. At th e same time, it even optimizes the number of processors to some degree for it ha s not been solved before. The performance has been observed in a representative example compared with other existing scheduling schemes in terms of several valu able factors. The experimental results show that this algorithm is feasible.