To conduct the cooperative hunting by multi-robots, the robots not only need to take into account basic problems (such as searching, pursuing, and collision avoidance), but also need to design the optimal task allocation mechanism and construct an efficient hunting alliance in order to capture the evaders quickly. In this paper, various related properties between the robots and targets are considered, and the quantitative criteria are defined. An assistant decision matrix is structured to establish the dynamic alliance. Then a pursuing strategy based on biologically inspired neural network is introduced to realize the hunting process after forming the dynamic alliance. The experimental results of simulation show the efficiency of the proposed approach.