3D face recognition has gained favour in the scientific community and industry due to the rapid development and decreasing cost of 3D sensors, with various novel techniques for face recognition presented in recent years. In comparison to 2D face images, 3D face images contain more explicit information, which is very useful to manage pose and illumination problems. However, the field of 3D face recognition is yet to fully mature and become widely used in industrial or commercial communities, mainly because of high error in non-cooperative and uncontrolled scenarios—particularly in challenging conditions of occlusions and partial data. Further, many existing 3D face recognition techniques require a training stage in their approach, which can suffer dramatic performance drop or even fail to work if only one training sample per person is available to the system. Thus, the one training sample issue is an important factor hindering the performance of 3D face recognition systems. In this thesis, we propose several 3D face recognition approaches to address the above issues. In the first half of this thesis, we propose two low-level structural representations 3D polygonal line chains (3DPLC) and 3D directional vertices (3D2V) to encode and match 3D faces.