A　novel　selection　method　(PLS_VIP_CFFS)　was　presented,　which　combines　partial　least　squares　variable　importance　in　projection　with　heuristic　forward　search　(PLS_VIP_CFFS).　To　prove　the　advantage　of　PLS_VIP_CFFS,　it　was　applied　to　the　selection　of　feature　genes　using　lung　cancer,　colon　cancer,　gastric　cancer　and　leukemia　data　sets.　Then,　the　feature　genes　were　used　in　classifiers,　such　as　SVM　and　CLUSTER.　The　specificity　sensibility　and　balance　accuracy　of　the　classifiers　and　the　major　biological　function　of　the　feature　genes　were　provided.　The　results　show　that　PLS_VIP_CFFS　is　a　reliable　selection　technique.