Real-time gaze tracking would be a key function for smart consumer electronics in the near future. But most gaze tacking system relies on strong IR light to enhance illumination, thus this results in poor user experience. In this paper, we present a high speed eye tracking system which is operated in a normal lighting condition. In addition to some image enhancement steps, the new system includes a dynamic training/prediction stage to further improve the detection accuracy of gaze points. The experimental result shows that the system can achieve the accuracy of 0.76° and 1.43° for the horizontal and vertical coordinates of the detected iris center, respectively.