Eye trackers have been used in various fields such as eye movement-related brain activity analyses, visual attention analyses and gaze input interfaces. Recently, these trackers have been improved in both terms of accuracy and portability. In this study, we propose a new eye tracking method which enables to estimate gaze direction from faces in images or movies without using the conventional eye trackers. The proposed method contains four steps: fitting a 3D model to the targeted facial image, extracting and geometrically correcting the regions of eyes using head-pose information, fitting a circle to each iris to localize its center, and compensating the POR corresponding to the head-pose changes. Our experiment’s result shows to measure its capability as gaze input interfaces based on accuracies of point-of-regards (POR).