Coherency identification is an important issue for transient stability analysis. In this paper, a coherency feature extraction method is proposed based on DFT-based continuous wavelet transform (CWT). By analyzing several typical situations of power angle swing (including incremental oscillated, damping oscillated and swing apart) using DFT-based CWT, it is illustrated that the scale and energy percentage of main components of the original signal can reveal the similarity and difference between power angle curves of the generators. Thus transient process of power angle can be described by a few indexes instead of a series of temporal data. Finally, case study in New England 10-machine 39-bus system indicates that the proposed coherency feature is valid for coherency identification in different fault cases.