Abstract: Emotion is an important aspect of human intelligence and is shown to play a significant role in creating virtual agent. Therefore, how to use emotion to enhance the intelligence and believability of virtual agent is an urgent issue to solve. Taken into account not only cognitive reasoning for generating emotion, but also physical interactions with its environment, a comprehensive computing emotional model is proposed. A complete architecture of emotion process is presented to describe the dynamic characteristics of emotional state and how to filter the mixture of emotions to generate coherent emotional state. The emotion structure with cause events is added to determine the specific effects the emotion will have on the decision making and behavior of the agent. In addition, a social learning component for predicting other agent's behaviours and relationships among agents is presented to improve virtual agents with the ability of adaptability in the dynamic environment. It is demonstrated experimentally through a computer simulation of virtual agents that the proposed emotional model can enhance the virtual agent's believability effectively.