In　this　paper,　we　analyzed　the　leukemia　gene　expression　profiles　based　on　the　method　of　bioinformatics,　and　focused　our　attention　on　the　leukemia　molecular　classification　and　informative　genes　identification.　After　having　removed　the　irrelevant　genes　to　the　classification　task,　we　employed　a　suboptimal　search　method　to　generate　candidate　feature　subsets　for　classification,　and　then　each　feature　subset　was　applied　to　support　vector　machine　to　classify　the　samples　by　＇Leave-One-Out　Cross　Validation＇　process　and　independent　test.　We　chose　the　genes　in　the　feature　subset　with　minimum　errors　as　the　informative　genes　for　distinguishing　the　two　classes　of　samples.　The　results　showed　that　all　the　samples　were　able　to　be　correctly　classified　with　informative　genes,　and　a　comparison　of　the　results　between　this　paper　and　some　previous　studies　was　also　presented.