To　improve　the　representability　of　emotion,　a　joint　feature　for　expression　sequences　by　combining　spatial　features　of　single　expression　image　frames　was　proposed.　On　the　basis　of　analyzing　the　representability　for　identifying　expressions　with　the　different　combination　of　rotations　and　scales　to　Gabor　wavelet,　the　Gabor　filter　with　three　rotations　and　two　scales　was　adopted　to　obtain　the　static　image　feature.　By　connecting　the　feature　of　series　expression　images,　the　joint　feature　was　established　by　containing　the　dynamic　property　of　an　emotion　action.　It　solved　the　problem　of　expression　description　by　using　the　expression　relative　local　spatial　information　and　the　sequential　change　clues　together.　Support　Vector　Machine　(SVM)　was　adopted　as　the　classifier,　and　the　test　was　done　on　the　JAFFE　static　expression　corpus　and　Binghamton　dynamic　expression　corpus.　Experiments　prove　the　effectiveness　of　feature　with　Gabor　and　PCA　comparing　to　PCA　only.　Results　also　show　that　the　joint　feature　based　on　dynamic　image　sequences　improve　the　expression　recognition　rate　referring　to　static　feature.