This paper describes the principle and examples of predicting users intent using the Bayesian method. In natural interfaces, users use continuous, undetermined multi-modal data to express their interaction intention rather than using discrete, determined actions. In order to interpret their interaction intend, we can either use “black-box” machine-learning methods or use “white-box” user-behavior-modelling methods. The essence of the latter is to computationally model the users interaction ability, which is crucial to understanding the users and exploring the computation principle of natural interaction. This paper summarizes the popular intelligent algorithms used in recent HCI researches, discusses the difference between these methods, and illustrates the methods of user modelling and the Bayesian method using some works from our laboratory.