Gaze tracking is a human-computer interaction technology, which has received extensive attention in academia and industry. Aiming at the problem of the large average error of the gaze points accuracy in non-invasive systems based on polynomial regression, a method of eye vectors calculation based on line-of-sight horizontal offset feature is proposed.On the basis of accurately extracting the coordinates of the iris center and inner corner of the eye on the face image, this method uses the inner corner coordinates as the reference point to calculate the offset of the iris center relative to the inner corner coordinates, calculates the weight of the horizontal eye movement vector through the offset, and then uses the weight to weight the horizontal eye movement vectors of the left and right eyes to synthesize new horizontal eye movement vectors. Using the data of 20 subjects with the head horizontal angle 0°from the Columbia Gaze database, this method was experimentally verified on 4 regression models. The results show that compared with the calculation method of binocular average eye vectors, this method can further reduce the horizontal of viewing angle error of the gaze points and improve the system accuracy.