Facial　Expression　Recognition　has　become　a　hot　research　direction　in　human　computer　interaction,　machine　learning　and　image　processing.　However,　the　inaccuracy　of　facial　expression　similarity　measure　is　always　a　key　problem　in　facial　expression　classification.　In　order　to　solve　this　problem,　a　novel　algorithm　which　is　based　on　Discriminative　Component　Analysis　algorithm,　is　proposed　in　this　paper.　Compared　with　Discriminative　Component　Analysis,　our　algorithm　is　based　on　maximizing　within-class　distance　and　minimizing　between-class　distance　to　choose　chunklets,　which　guarantee　the　stability　and　accuracy　of　algorithm.　The　mean　recognition　rates　are　80.42%　and　95.71%　under　the　condition　of　using　hold-out　method　and　leave-one-out　method　respectively.　Experimental　results　show　the　effectiveness　of　our　algorithm.　©　2012　IEEE.