For the shortcoming of previous fault diagnosis studies that the upper of faults or faults derivative and disturbance of system should be known, as well as the difficulty to diagnose the actuator faults and sensor faults simultaneously, an adaptive unknown input fault diagnosis observer is proposed, which can reconstruct the actuator faults and sensor faults in nonlinear dynamical system simultaneously. Firstly, an H∞ performance index is employed to restrain the influence to fault reconstruction caused by the unknown input, and a Lyapunov function is employed to obtain the robust asymptotically stability of the observer error dynamical system. Then, the gain matrices of the observer are solved by using the linear matrix inequality, and fault reconstruction is fulfilled. Finally, results of the simulation on a DC motor system show the effectiveness of the proposed method.