Aiming at the issue of analog circuit fault diagnosis and location, a novel approach for analog circuit fault diagnosis based on continuous wavelet Tsallis singularity entropy (TSE) and extreme learning machine (ELM) is proposed to enhance the accuracy of fault diagnosis. Firstly, the fault response signals are preprocessed by the continuous wavelet transformation to obtain the timefrequency coefficient matrix, and the matrix is divided into 8 congruent timefrequency blocks. Then, the feature vector is obtained by computing TSE of each block. Finally, the feature vectors are used as the inputs of a kind of multiclass classifier, namely ELM. The simulation results demonstrate that the proposed fault diagnosis approach can not only extract the essential features of fault response signals with better performance, and also achieve higher diagnosis accuracy than other reported approaches.