A　method　based　on　support　vector　machines　is　proposed　for　cancer　molecular　classification.　The　Bhattacharyya　distance　of　each　gene　is　calculated　as　the　criterion　for　filtering　＇noisy-genes＇　which　do　not　contribute　to　classification,　and　then　based　on　the　expression　values　of　the　informative　genes,　a　linear　support　vector　machine　is　used　as　the　classifier　to　classify　different　cancers.　This　process　has　been　applied　to　the　human　acute　leukemia　dataset　as　a　test　case.　The　effectiveness　and　feasibility　of　the　method　is　proved　by　the　results　that　all　the　samples　in　the　dataset　can　be　correctly　classified　without　any　error.