For　large　data-set,　the　speed　of　algorithms　for　support　vector　machines　is　one　restriction　for　their　performances;　besides,　noise　and　outliers　which　the　data-set　contained　also　influenced　their　capabilities　greatly.　This　paper　proposes　an　effective　method　for　weighted　support　vector　regression　(SWSVR):　the　simplification　of　the　original　samples　(contain　noise　and　outliers)　is　taken　firstly;　then,　the　selected　samples　are　trained　by　the　weighted　support　vector　regression　machine.　The　results　of　the　experiment　shows　that　our　method　not　only　speedups　the　calculation,　but　also　meliorates　the　performance　of　the　regression.