Traditional　damage　detection　methods　for　nonlinear　structures　are　often　based　on　simplified　models,　such　as　the　mass-spring-damper　and　shear-building　models,　which　are　insufficient　for　predicting　the　vibration　responses　of　a　real　structure.　Conventional　global　nonlinear　finite　element　model　updating　methods　are　computationally　intensive　and　time　consuming　Thus,　they　cannot　be　applied　to　practical　structures.　A　decentralized　approach　for　identifying　the　nonlinear　material　parameters　is　proposed　in　this　study.　With　this　technique,　a　structure　is　divided　into　several　small　zones　on　the　basis　of　its　structural　configuration.　The　unknown　material　parameters　and　measured　vibration　responses　are　then　divided　into　several　subsets　accordingly.　The　structural　parameters　of　each　subset　are　then　updated　using　the　vibration　responses　of　the　subset　with　the　Newton-successive-over-relaxation　(SOR)　method.　A　reinforced　concrete　and　steel　frame　structure　subjected　to　earthquake　loading　is　used　to　verify　the　effectiveness　and　accuracy　of　the　proposed　method.　The　parameters　in　the　material　constitutive　model,　such　as　compressive　strength,　initial　tangent　stiffness　and　yielding　stress,　are　identified　accurately　and　efficiently　compared　with　the　global　nonlinear　model　updating　approach.