Aiming at the agents group composed of multiple leader agents and multiple follower ones, a flocking behavior method with multiple leaders in obstacle environment is proposed. As a distributed control method, it controlled the follower agents with the leader agents of the group. Simultaneously, the center and average velocity of the follower agents exponentially was converged to the weighted average position and the velocity of the leader agents respectively. And the weight was dependent on the number of follower agents which were led by the leader agents. Moreover, by introducing a class of agents which are denoted as “β-agent”at the border of the obstacles, the follower agents possessed the capacity of avoiding collision with the static and dynamic obstacles. In this way, believable flocking behavior with multiple leaders in obstacle environment was generated. Simulation results with virtual humans being the members of the group demonstrate the effectiveness of the proposed method.