Differing　in　traditional　methods　which　extracted　too　much　features　or　filtered　valuable　items,　we　proposed　a　feature　representation　and　extraction　method　based　on　LZW　compression　algorithm　to　detect　malicious　codes.　The　compression　algorithm　not　only　reduces　the　number　of　features,　but　also　is　enough　to　cover　malicious　codes.　In　this　paper,　we　described　the　process　of　our　feature　extraction　in　detail,　including　0-data　processing,　fix-length　coding　and　threshold　setting.　The　experimental　results　show　that　our　method　outperforms　other　methods　based　on　Bayes　and　SVM　in　DR　and　AR.　©　2013　Springer-Verlag　GmbH.