A　new　batch　least　square　(LS)　identification　algorithm　with　recursive　formulation　is　proposed　to　deal　with　the　problems　occurred　when　the　ordinary　LS　algorithm　is　applied　to　real-time　and　on-line　identification.　The　fundamental　feature　of　this　algorithm　is　to　fix　the　observation　matrix＇s　dimension,　making　use　of　definite　amount　of　newly　obtained　data　and　ignoring　the　older　data.　Taking　advantage　of　recursive　formulation,　this　algorithm　avoids　the　inverse　operation　and　thus　reduces　the　calculation,　making　itself　capable　of　application　in　real-time　and　on-line　identification　case.　Simulation　shows　that　this　algorithm　has　rather　good　precision　and　tracking　ability,　which　are　greatly　influenced　by　the　dimension　of　the　observation　matrix.