Abstract: Uncertainty diagnosis and reconstruction of the nonlinear system based on multi-sliding mode observers is proposed for a class of nonlinear system with uncertainty. The method divides the original system into two subsystems via nonlinear transformation, and the multi-input external disturbance can be eliminated by multi-sliding mode. The unknown input can be guaranteed to be observed and reconstructed by the design of the multi-sliding mode observers under hypothetical construct of the disturbance distribution matrix and the unknown inputs can be reconstructed via the outputs estimated error on the multi-sliding mode. Simultaneously, the estimation feedback gain design assures the asymptotical stability of the residual errors. Finally, the effectiveness of the proposed method is demonstrated by a simulation example on a single-link flexible joint robot.