At present stage, active power oscillation of electric system can be classified into two categories—negative damping and forced power oscillation according to its generation mechanism. Although the two power oscillation modes are similar, their control measures needs to be adopted are totally different. Therefore, distinction of the two oscillation types becomes precondition for suppressing the oscillation with proper measures. This paper proposes a practical approach to recognize oscillation types by identification of real-time power oscillation curves recorded by WAMS. Hilbert-Huang transform is employed to obtain envelope curve of power oscillation curve, based on which 100 sampling points are selected to train and test the neural network supporting vector machine. A supporting vector machine for identifying the characteristic of power oscillations is trained by simulation data of 16 machines 68 nodes power system. Then, this paper applies this supporting vector machine to identify the power oscillation curves from 16 machines 68 nodes power system and real power grid. All tests indicate that the proposed oscillation recognition method possesses good precision and is provided with practical engineering application. This work is supported by National Natural Science Foundation of China (No. 51177057).