Abstract: With the expansion of control systems in scale, the possibility of fault is getting high. The security and reliability of a control system becomes a key issue. So far, the fault diagnosis for linear systems sees many successes, but that for a nonlinear system remains a difficult problem. The traditional fault diagnosis method based on state observer can not effectively used for nonlinear systems; and its robustness of the diagnosis and localization can not be well ensured when a system has uncertainties and disturbances. In this paper, for the sensor fault diagnosis and for nonlinear systems, a method for fault detection and reconstruction based on sliding mode observers is proposed. The sensors fault is first augmented into actuator faults by introducing integral variants. Then a sliding mode observer is constructed for the augmented system. The stability of the sliding observers is proven by using the Lyapunov stability theory. Finally, the observers are designed by solving the Linear Matrix Inequalities (LMI) and the sensor faults can be reconstructed directly. Numerical simulation shows the effectiveness and accuracy of the reconstruction of the sensor faults using the proposed approach.