According　to　the　fact　that　the　bootstrap　in　SVM　ensemble　learning　can＇t　generate　the　committee　classifiers　with　big　differences,　SVM　ensemble　using　bisecting　grid-based　method　is　proposed(GBSVME).　By　hierarchically　bisecting　each　grid　into　two　volume-equal　new　grids,　this　approach　use　a　new　criterion　to　measure　the　significance　among　all　grids.　Then,　using　a　random　method　to　select　some　important　grids　to　be　further　bisected.　Therefore,　the　proposed　approach　can　divide　all　data　into　some　grids,　and　use　all　the　grids　as　the　input　for　training　committee　SVMs.　Two　experimental　results　show　that　the　performance　of　GBSVME　is　better　than　that　of　mang　other　ensemble　algorithms.　©　2011　IEEE.