In order to solve the problem in the feature extraction of analog circuit, a method based on Treelet transform is proposed in this paper. Treelet transform is an adaptive multi-scale data analysis method that can be used in the high dimensional data dimension reduction and feature selection.This approach performs Treelet transform on the acquired output signals of the test circuit as the fault feature vector.Then the obtained feature vector is used as the input of the BPNN to recognize the fault patterns. The simulation result showed that the proposed approach can effectively extract the fault feature vector. Compared with other feature extraction methods, Treelet transform has good accuracy of analog circuit fault diagnosis.