Abstract: It is important to locate the center of distorted pupil in eye tracking process. To tackle the issues induced by distortion and disturbance, a new Random Sample Consensus algorithm based on only 3 points, named RANSAC_3, is proposed in this paper. Ellipse model is adopted to match the distorted pupil, its parameters are calculated from the two randomly sampled points and one searched point. Only two randomly sampled points are needed to estimate the ellipse parameters in this method, so the probability of sampling disturbing points is reduced. The point derived by searching can be used to determine whether to calculate the ellipse parameters, so the efficiency of ellipse fitting and the accuracy of localization are improved. Experimental results show that the method has pretty good performance on the localization of distorted pupils, and is effective and robust to various disturbances, i.e. spectrum reflection, eyelash, hair, glasses and motion blur. Moreover, its high speed ensures that it can be used in a real-time eye tracking system.