This paper presents a binocular eye tracking device which used 2-D regression-based method to avoid the complicated vector calculations, often applied in geometry-based 3-D tracking, to increase processing speed. It is probably due to the speed increasing that enables us to do the experiments of dynamic estimation errors, which are actually like an experiment to trace a moving object in real 3-D environment if the predefined points, used for evaluating the error estimation, increased to a large number. So far, the preliminary estimations seemed acceptable. It also found 3-D estimation errors mainly caused by the 2-D calibration errors. Therefore, seeking a better 2-D calibration would become one of the important issues for further experiments.