A　new　online　self-constructing　recurrent　neural　network　(SCRNN)　model　is　proposed,　of　which　the　network　structure　could　adjust　according　to　the　specific　problem　in　real　time.　If　the　approximation　performance　of　SCRNN　is　insufficient,　SCRNN　can　create　new　neural　network　state　to　increase　the　learning　ability.　If　the　neural　network　state　of　SCRNN　is　redundant,　it　should　be　removed　to　simplify　the　structure　of　neural　network　and　reduce　the　computation　load;　otherwise,　if　the　hidden　neuron　of　SCRNN　is　significant,　it　should　be　retained.　Meanwhile,　the　feedback　coefficient　is　adjusted　by　synaptic　normalization　mechanism　to　ensure　the　stability　of　network　state.　The　proposed　method　effectively　generates　a　recurrent　neural　model　with　a　highly　accurate　and　compact　structure.　Simulation　results　demonstrate　that　the　proposed　SCRNN　has　a　self-organizing　ability　which　can　determine　the　structure　and　parameters　of　the　recurrent　neural　network　automatically.　The　network　has　a　better　stability.　©　2011　Springer-Verlag.