针对智能虚拟环境中决策所必须解决的实时性和动态性的问题，研究了智能虚拟环境模型的构，对传统的规划方法做了,要用了能进行实时规划的限时A＊算法，同时使用了动作栈，该方法可以对当前的动作进行即时评价，使得决策可以满足动态开放的虚拟环境要求，在此基础上实现了一个决策Agent，它能产生一定的智能行为，最后实现了一个小型的智能虚拟环境，并应用该模型来指导其中虚拟人物的行为。

This paper discusses the structure of decision model in intelligent virtual environment. Intelligent virtual environment is dynamic environment, that is to say it keepes on changing, decision in intelligent virtual environment also should be dynamic, it must be rapid enough to react to the change of environment. To meet the real time and dynamic requirement of decision in intelligent virtual environment, this paper improve the traditional planning method. The time critical A * algorithm is used to implement real time planning, current action is generated only in a predetermined decision time slice, which is different to traditional planing method. action stack is also employed to store the selected action, which is convenient to evaluate the action in time according to the situation of environment, so, we can discard those unreachable targets in time. Based on that, this paper developes a decision agent to instruct the virtual human in intelligent virtual environment. In addition, a small intelligent virtual environment is developed to test the decision model. At last, Influence of length of decision time slice is test, it shows that the suitable length of decision time slice is about 0 3 seconds.