The problem of dynamic task allocation for autonomous underwater vehicles(AUVs) has high nonlinearity, uncertainty and multi-modal characteristics, which requires that the task allocation method has self-organization, robustness and rapidity. The dynamic ant colony's labor division(DACLD) mechanism is a group intelligence algorithm applying a distributed framework. The overall intelligent behavior that emerges from many simple interactions can be well adapted to complex and variable environment, which is very flexible in solving the problem of task allocation. This paper presents a multi-AUV task allocation method based on DACLD, in which the stimulus-response principle of ant colony division of labor is introduced and the mapping relationship between ant colony division of labor and multi-AUV task assignment is established, and the task state estimation is incorporated into the response threshold. And also, a new cyclic competition scheme is proposed to maximize the utilization of AUV resources for the task conflict phenomenon. The simulation results show that the proposed method can efficiently complete the task allocation process and has good self-organization, robustness and rapidity.