In　this　paper,a　noninvasive　method　for　Knee　Osteoarthritis　(KOA)　detection　and　diagnosis　is　proposed　using　data　from　surface　electromyogram　(sEMG)　signals　with　the　purpose　of　accessing　the　state　of　KOA　in　the　early　stage.　In　our　experiment,　sEMG　are　collected　from　rectus　femoris,　vastus　medialis,　biceps　femoris,　semitendinosus　muscle　of　control　group　and　KOA　group　respectively　when　they　are　in　the　walking　model,　then　parameters　of　autoregressive　recurrent　model　(ARM)　based　on　which　are　extracted　by　the　well-known　Kalman　filter　as　the　characteristic　vectors,　which　is　used　to　train　the　RBF　neural　network.　Finally,　the　knee　osteoarthritis　will　then　be　diagnosed　through　the　RBF　neural　network.It　is　shown　that　a　much　improved　result　over　the　traditional　method　is　achieved　over　classifiers　based　on　RBF　neural　network.　©　2012　IEEE.