A　set　membership　identification　method　by　pattern　classification　is　proposed　for　nonlinear-in-parameter　regression　models　with　unknown　but　bounded(UBB)　noises.　Suppose　that　the　points　in　the　parameter　space　can　be　divided　into　two　classes　according　to　whether　they　are　in　the　feasible　solution　set　or　not,　the　problem　of　set　membership　identification　is　to　construct　a　pattern　classifier　to　decide　which　class　a　point　belongs　to.　The　method　has　three　steps.　Firstly,　the　training　data　are　selected　uniformly　in　the　parameter　space　and　are　decided　by　equation　error　whether　they　are　in　the　feasible　solution　set.　Secondly,　supervised　locally　linear　embedding(SLLE)　is　used　to　map　the　training　data　into　low-dimensional　space.　Thirdly,　nearest　mean　classifier(NMC)　is　trained　on　the　mapped　training　data.　This　method　not　only　can　describe　the　feasible　solution　set　approximately　in　the　high-dimensional　parameter　space,　but　also　can　characterize　it　in　the　low-dimensional　feature　space.　Simulation　results　show　the　effectiveness　of　the　proposed　method.