Abstract: A design method of unknown input observer which is able to estimate simultaneously the system states and the unknown inputs for affine nonlinear systems with unknown inputs is discussed. First, the affine nonlinear system is linearized by using Lie derivative and transformed into a controlled normalized form. Second, for the transformed equivalent system, a higher-order high-gain sliding mode observer, which is able to estimate not only the system states but also their state differentials, is developed. Furthermore, a way to reconstruct the unknown inputs is proposed based on the estimated states and their differentials. The proposed method can reconstruct large time-varying unknown inputs while system differential information isn't reqired. Finally, two simulation examples on real models are given to verify the effectiveness of the proposed method.