Reconstructing the 3D facial model of an unidentified individual from his skull contributes considerable benefits to terms of archaeology, anthropology and forensic investigation but it is still significantly complicated matter. Computer aided system of 3D facial reconstruction based on skull has a great advantage of reduction in time consumption. Nevertheless, existing results either contain graphical artifacts or reflect incorrectly differences among reconstructed faces. In this paper, we propose a Vietnamese 3D facial reconstruction system from skull that can overcome existing problems. Machine learning is applied to CT image database of living Vietnamese to learn the relationship between skull measurements and soft tissue thicknesses. Then, the soft tissue thicknesses for each individual are calculated from his/her skull measurements. Moreover, instead of using a fixed number of dowels, extra dowels are added to deform a 3D template face, which ensures the smoothness of the 3D reconstructed face as well as represents best the characteristics of the face to be reconstructed.