Based　on　linear　combination　model,　a　new　facial　feature　extraction　method　is　proposed.　By　matching　procedure,　the　combination　model　extracts　automatically　the　feature　points　of　given　facial　image　from　the　information　of　the　labeled　prototypes.　To　construct　linear　combination　model,　a　restrained　optical　flow　algorithm　is　proposed　to　solve　the　difficult　problem　of　the　pixel-wise　alignments　among　the　labeled　prototypic　faces.　By　implementing　the　feature　extraction　method　on　the　MPI　and　ORL　face　database,　we　gain　average　error　96.7%　and　86%　respectively.　The　result　shows　that　the　method　has　good　performance.