The extraction of fault features is one of the key technologies in analog circuit fault diagnosis. To acquire effective features, a method to extract the fault features based on fractional Fourier transform (FrFT) and fractal dimension (FD) is proposed. The original feature data is mapped to different fractional space and the FD is computed. Furthermore, the FrFT-FD feature is carried out for data dimensionality reduction by using kernel principal component analysis (KPCA). Finally, the optimized feature vector is diagnosed by neural network (NN). The simulation results show that the proposed method can acquire a subtle difference, enhance the separability of different fault modes, and improve the diagnostic accuracy.