The invention discloses a pixel-level precision human eye fixation point positioning method. The method comprises the steps of calculating the gradient of an eye region image after passing through a key point of a depth network, and further correcting the central position of a pupil. Compared with an existing sight line estimation method, the pupil center position can be positioned more accurately, and especially under the condition that head or eyeball offset is large. According to the embodiment of the invention, the sight line estimation precision can be effectively improved, and the position of an implementation fixation point of a user is further positioned. In addition, the distance from the eyes to the screen is estimated through the deep network, and then the fixation point of theuser on the screen is estimated. And compared with a pupil corneal reflex method, only a single network camera is adopted, so that the equipment cost is greatly reduced. Compared with an existing single-image processing method, the method does not need to limit the posture of the head, and the robustness of the algorithm is greatly improved. Through matching with the 3D face model, the limitationproblem that an existing database cannot represent all postures is solved, and the practicability of the method is improved.