Abstract: To overcome the imprecision of fault feature extraction for analog circuits because of the mixing phenomenon of wavelet analysis,a method of fault feature extraction for analog circuits based on empirical mode decomposition(EMD) is put forward.The method gets several intrinsic mode functions(IMF) through the EMD of analog circuit output signals,and the energy of each IMF is regard as the feature to distinguish faults.Aiming at the problem of end effect on EMD,a predictive method of least square support vector machines(LSSVM) based on genetic optimization is put forward.The results of simulation experiment prove the validity of this method.