In a method for facial feature point tracking, facial feature points of a face in a previous image of a to-be-tracked image are acquired in an image sequence. Facial feature point errors between the to-be-tracked image and the previous image are determined based on a preset error model and pixels in the to-be-tracked image. The facial feature point errors indicate differences between first coordinates of facial feature points in the to-be-tracked image, and second coordinates of facial feature points at corresponding positions in the previous image. The preset error model is trained based on facial feature points in a plurality of pairs of adjacent reference images. Further, the facial feature points of the face in the to-be-tracked image are determined based on the facial feature points of the face in the previous image and the facial feature point errors between the to-be-tracked image and the previous image.