Abstract: We present a hierarchical behavior model to simulate realistic crowd behaviors. This model is composed of two parts. One is the low density behavior module, emphasizing the autonomy and diversity of the behaviors. The other is high density behavior module, paying attention to showing the flocking instincts of the crowd. We design a new mechanism to schedule them to generate appropriate behaviors based on the crowd density. We also take into account the impact of the crowd density on the motion speed and provide an interface for the users to control the motions of the crowd. The experimental results show that our hierarchical behavior model not only well reflects the characteristics of the human behaviors, but also shows plausible crowd behaviors to respond properly to certain event.