Aiming at the problem of parametric fault diagnosis in nonlinear circuits, an approach utilizing cepstrum and decision tree is proposed. Firstly, the acquired fault response signals are converted by cepstrum. Then, the wavelet analysis is used to decompose the converted data and the energy is taken from different frequency bands. Finally, the obtained fault features are inputted into decision tree to identify different faults. The simulation results show that the proposed method can extract the fault signature effectively and can get a good diagnosis result.