The work focus on the problem of tracking faces and facial actions based on Candide3 facial model. The main contribution of this paper is described as follows. First, a semi-automatic method is developed to initialize face model parameters. Second, an example-based learning method is utilized based on preliminary sample generating work using the semi-automatic method. Third, in order to restrain the effect of illumination, a normalization method is applied. Through the simulation experiments, a conclusion can be summarized that the method in this paper is more convenient and accurate.