Recently, detection of structural damage based on the system identification has received great attention. In this paper, a technique is proposed for the identification of nonlinear structural parameters under unmeasured excitation. The identification algorithm is based on the extended Kalman filter for the extended state vectors including nonlinear parameters and the recursive least squares estimation for the unknown inputs. Two different models are used to simulate nonlinear structures: One is a 4-storey Duffing-type nonlinear elastic shear-frame structure, the other is a 4-storey Bouc-Wen hysteretic shear-frame structure.Two numerical examples are carried out on the two kinds of models. The simulation results demonstrate that the proposed approach is capable of identifying the nonlinear structural parameters and unknown inputs with good accuracy.