Abstract: Based on the analysis of the algorithms commonly used for task assignment, this paper puts forward an improved both-initiated adaptive algorithm for task assignment. This algorithm adaptively determines initiated strategy according to the system load. In the system, the data requirement of each task and the data distribution are considered.The nodes are effectively chosen so as to minimize the costs for task assignment. This algorithm has been successfully applied in distributed parallel database system DPSQL. It has been shown that the method has better efficiency and stability than the common algorithms for task assignment.