Generalized　hyper-sphere　SVM　is　a　promising　method　for　the　pattern　classification.　The　ratio　of　the　support　vectors　from　two　classes　of　samples　can　not　be　adjusted　conveniently　by　setting　the　parameters　η　and　b　in　the　generalized　hyper-sphere　SVM　(GHSVM),　which　affects　the　generalization　performance　to　some　extent.　A　weighted　hyper-sphere　SVM　is　studied　in　this　paper.　The　results　shows　that　the　margin　may　be　obtained　much　more　easily　by　weighted　method　rather　than　by　adjusting　the　parameters　n　and　b,　which　makes　the　classifier＇s　generalization　performance　much　better　than　the　original　GHSVM.