The existing methods that are used for face recognition from 2D images do not always provide an opportunity to gain intermediate information about the proportions of a human face, these algorithms are also very sensitive to lighting, which decreases the accuracy of face recognition. Based on that there is a need of new methods for reconstructing a 3D model of a face to generate new images of a face (turn, tilt of the head). In this paper, a method is proposed for adapting an average polygonal 3D face model from several 2D images.