Actual　water　quality　monitoring　sites　due　to　inadequate　and　abnormal　by　the　impact　to　water　quality　warning　has　brought　new　challenges.　How　do　based　on　limited　monitoring　data　to　predict　the　water　quality　to　address　the　shortage　of　river　water　quality　monitoring　sites　and　data　coverage　of　false　alarms　caused　by　abnormal,　early　warning　of　water　pollution　are　of　great　significance.　According　to　river　water　monitoring　data,　the　small　sample　properties,　is　proposed　based　on　least　squares　support　vector　machine　prediction　of　water　quality　parameters,　this　method　has　strong　ability　to　predict　the　true　value,　and　the　global　optimization　and　good　generalization.　This　method　is　applied　in　the　river　water　quality　measurement　data,　after　training　the　LS-SVM　model　for　water　quality　parameters　of　water　quality　monitoring　system　to　predict,　in　the　same　sample　under　the　BP　network　and　RBF　network　prediction.　Experimental　results　show　that　the　small　sample　case　with　noise,　least　squares　support　vector　machine　method　is　better　than　multi-layer　BP　and　RBF　neural　network,　to　better　meets　the　requirements　of　water　quality　prediction.　Comparing　data　of　the　two　experiments　shows　that　new　module　makes　the　interest　mining　more　effective.　(C)　2011　Published　by　Elsevier　Ltd.　Selection　and/or　peer-review　under　responsibility　of　Kunming　University　of　Science　and　Technology