Abstract: The nonlinear restoring force (NRF) after earthquakes can be used to directly describe the initiation and development of structural damages and the variation of structural damage pattern. In this study, a nonparametric identification approach for structural nonlinear restoring force, mass and seismic acceleration using partially available measured dynamic responses and partial mass is proposed and is numerically validated with multi-degree-of-freedom (MDOF) structure equipped with nonlinear members. Firstly, the time history of ground motion is identified using dynamic responses measurement and mass information at limited degrees of freedom (DOFs). Secondly, the unknown structural mass and the time history of NRF are identified with an extended Kalman filter with memory fading technique and weighted global iteration method (EKF-MF-WGI) in a nonparametric way, where a power series polynomial (PSP) is employed to model structural NRF. Numerical study with a MDOF nonlinear structure equipped with a Magnetoreheological (MR) damper is carried out considering measurement noises and different initial mass estimation errors. The results show that the structural mass, NRF in time domain and seismic acceleration can be identified effectively by the proposed approach.