Aiming　at　several　common　drawbacks　of　existing　gaze　tracking　algorithm　which　is　based　on　image　processing　technology:　low　accuracy,　low　suitability　and　high　requirement　of　device,　this　paper　presents　an　improved　gaze　tracking　method　based　on　the　human　eye　model.　The　vision　scatter　model　is　established　by　studying　the　relationship　among　the　center　of　iris,　inner　corner　of　eye　and　the　gaze　direction.　The　model　is　used　for　gaze　rough　estimation.　An　improved　geometric　eye　model　with　compensation　coefficient　is　used　for　gaze　precise　estimation,　and　the　coefficient　depends　on　the　result　of　rough　estimation.　The　result　of　experiment　shows　that　the　proposed　approach　has　a　low　average　error　which　is　reached　2　°　and　good　robustness.　This　approach　provides　an　effective　solution　for　gaze　tracking　of　human-computer　interaction　under　the　low　pixel　condition.　©　2014　TCCT,　CAA.