Abstract: For problems that fault feature extraction was not enough and fault line selection criterion needs samples with long training time in existing fault line selection methods for distribution network, a fault line selection method for distribution network based on measurement of characteristic atom was proposed. Firstly, cosine packet and db10 wavelet are used to construct steady-state component atom and transient component atom of transient zero-sequence current, so as to form mixed atom dictionary, therefore characteristic atom of each line is extracted from the mixed atom dictionary by use of matching tracing algorithm. Then, in the aspect of energy, energy entropy is used to measure energy of the characteristic atom, and in the aspect of waveform similarity, correlation coefficient between the characteristic atom and the original transient zero-sequence current is calculated to measure similarity of waveforms, so as to get modified atomic energy entropy of each line. Lastly, comprehensive atomic energy entropy which represents physical characteristic of the each line is obtained by differential operation of the modified atomic energy entropy, and the line with the largest comprehensive atomic energy entropy is judged as fault line. The simulation results show that the method is accurate and effective with strong anti-noise interference ability, which is not affected by transition resistance, number of iterations, length of data window, compensation degree of arc suppression coil and capacitor switching.