In the view of the systems model uncertainties, the unknown input disturbances and the non-linear characteristics of the system model, the fault diagnosis algorithm of a class of nonlinear system based on parameter estimation is proposed. A fault diagnosis observer is constructed for the systems considered, and the Lyapunov stability theorem is used to verify that the fault diagnosis observer is steady. The fault diagnosis observer that characters the parameter fault systems converges asymptotically through the Barbalat lemma. Finally, the algorithm procedures are concluded. The simulation results show that the proposed algorithm has the characteristics that it converges quickly, and then has a perfect diagnosis effect on a class of nonlinear systems.