For the traditional matrix pencil algorithm, it is difficult to accurately identify the signal parameters and the error is bigger under low SNR. Therefore, a new method about dominant mode identification of low frequency in power systems is proposed, which combines ensemble empirical mode decomposition (EEMD) and matrix pencil algorithm. The signal is smoothed by using EEMD, and the cross-correlation coefficients and signal energy weighting ratio are used to select IMF which has main mode, and the above mode parameters are obtained by matrix pencil algorithm, which enlarge the scope of traditional matrix pencil algorithm. Analysis results show that, with adaption to nonlinear system and strong anti-noise ability, the method still can be used to identify the dominant mode of low frequency oscillation, providing a new idea to study low frequency oscillation in power system. This work is supported by National Natural Science Foundation of China (No. 51577053).