Time-frequency atom decomposition has good analysis effects on power quality disturbance signals. However, commonly-used matching pursuit (MP) algorithm has several defects, such as the heavy burdern of computation, the impacts of spacial discretization on the performance, etc. An in-depth study is conducted on the evolutionary matching pursuit (EMP) of power quality disturbance signal atom decomposition based on differential evolution (DE) algorithm. The algorithm steps are presented in detail. For several typical power quality disturbance signals, the independent simulation experiments of 30 times are implemented with both Gabor and damped sinusoid atoms. An analysis is made on the performance impacts of the signal length and the noise. The simulation results show that, compared with the MP algorithm, the EMP algorithm greatly reduces the computation time without the influence from the signal length. Moreover, the global matching performance of the optimal atoms is further improved without the influence from the noise. Finally, the future research prospects are introduced.