selecting　a　subset　of　marker　genes　from　thousands　of　genes　is　an　important　topic　in　microarray　experiments　for　diseases　classification　and　prediction.　The　SVM-RFE　is　popularly　employed　to　select　feature.　In　this　paper,　we　proposed　a　hybrid　approach　to　select　marker　genes　for　tumor　classification.　Firstly,　filter　method　was　employed　to　selected　informative　genes,　and　then　we　improved　the　standard　SVM-REF　to　extract　feature　genes　from　the　small　set　of　informative　genes.　The　improved　SVM-RFE　accelerates　without　reducing　accuracy　the　standard　support　vector　machine　recursive　feature　elimination　method.　Our　method　has　been　implemented　on　ALL/AML　dataset,　and　the　results　have　shown　that　our　method　can　achieve　to　select　few　of　marker　genes　with　minimum　redundancy　but　getting　better　classification　accuracy.