The stability requirement of the densely interconnected power system indicates the importance of low-frequency oscillation mode identification. Eigensystem Realization Algorithm (ERA) is a time-domain technique for parameter estimation. It is capable of extracting information from multiple-channel input signals even under noisy conditions. However, ERA requires the prior knowledge of the number of modes present in the signal. Therefore, a modified ERA-based mode identification method is proposed in this paper with a new way of order determination. It is demonstrated that the proposed method can obtain an accurate model order estimation at different noise levels. Furthermore, the proposed method is tested on a 36-bus benchmark system as well as on measurement data from a real power system. Comparison of the proposed method with other methods is also carried out. The simulation results confirm the efficacy of the proposed method.