In order to efficiently test and high speed diagnose analog circuits, a new analog circuit fault diagnosis method based on LMD multiscale entropy and extreme learning machine is proposed in this paper. First, the fault signal is decomposed into several production functions by LMD. Then, the multiscale entropy of each PF component is worked out and fault feature vectors are constructed. Finally, the fault feature vectors are feed into the extreme learning machine to train and test. The simulation results show that the diagnosis time only needs 0.028 74 s, and the accuracy of fault diagnosis can achieve 98.89%. Compared with other three ways, the method can effectively reduce the diagnosis time and improve the accuracy of fault diagnosis.