Based　on　discriminative　component　analysis　(DCA)　algorithm,　a　local　discriminative　component　analysis　(LDCA)　algorithm　for　facial　expression　recognition　is　proposed.　First,　LDCA　algorithm　chooses　a　number　of　nearest　neighbors　of　a　test　sample　from　a　training　set　to　capture　the　local　data　structure.　Then,　the　facial　expression　features　of　each　testing　sample　are　extracted　by　maximizing　the　total　variance　between　the　discriminative　data　chunklets　and　minimizing　the　total　variance　of　data　instances　in　the　same　chunklets.　The　experimental　results　on　several　representative　facial　expression　datasets　show　that　proposed　method　not　only　improves　the　recognition　rate　of　DCA　algorithm,　but　also　exhibits　strong　robustness.