The multi-target, multi-task assignment problem for multi-agents systems is studied. By considering the precedence relationships between the tasks, the model of the distributed task allocation for agents is established. The consensus-based bundle algorithm(CBBA) is extended by distributing tasks to different layers according to its precedence. Targets are only allowed to add to the bundle and path of an agent when its precedence task has been assigned, so that the precedence constrains are satisfied, and the properties of CBBA are preserved. The simulation of comparing verifies the proposed task planning methodology for complex missions.