In　this　paper,　we　describe　a　combination　model　for　3D　facial　synthesis.　To　construct　the　model　it　is　fundamental　to　compute　the　dense　point-to-point　alignments　among　the　prototypic　faces　acquired　by　Cyberware　scanner.　The　proposed　mesh　resampling　method　overcome　the　alignments　by　the　same　resampling　criterion　to　all　prototypes.　From　the　aligning　prototypes　the　linear　combination　facial　model　is　constructed.　Matching　the　model　to　given　facial　image,　the　3D　face　will　be　built　automatically.　To　accelerate　the　model　matching　and　expand　the　model　fitting　complex　illumination,　the　multi-resolution　model　and　multi-light　model　are　designed.　Although　sharing　the　basic　idea　with　the　Morphable　Model,　the　combination　model　has　innovations　in　alignment　strategy　and　illumination　model.　The　experimental　results　show　that　the　combination　model　have　good　performance　in　3D　face　modelling.