Abstract: In view of the current eye tracking methods were too sensitive to conditions that face scale variations, partial occlusion of eye and head rotations in horizontality and verticality, which often resulted in tracking failure for the loss of objects. One eyes tracking method was proposed that was based on Kalman filter and improved Camshift algorithm. Firstly, predict the position of the eyes in the current image by Kalman filter; secondly, search the eyes region by Camshift iteration algorithm at the center of predicted position; thirdly, adjust the search window in the light of the symmetry of the of eyes; finally, update Kalman filter and the eyes templates. The experimental results show this method is robustness to those proposed conditions.