A new method about dominant low frequency oscillation mode identification for power system is proposed. The method combines Multivariate Empirical Mode Decomposition (MEMD), Teager energy operator and Prediction Error Method (PEM). The signal containing low frequency oscillation characteristic information of power system is decomposed by MEMD into several Intrinsic Mode Functions (IMFs) components. The main IMF containing dominant oscillation mode is selected by means of the rapid response ability of Teager energy operator. The Prediction Error Method (PEM) is applied to identify the oscillation frequency and damping of each dominant oscillation mode. The method is validated by IEEE68 bus test system and the PMU data of the Liaoning Power Grid. The analysis results demonstrate that this method can effectively identify the dominant oscillation mode from the wide-area measurement information of power system. This work is supported by Science and Technology Project of State Grid Liaoning Electric Power Company (No. SGTYHT17JS201) and Science and Technology Project of State Grid Corporation of China (No. SGTYHT17JS199).