There is a wealth of dynamic information hidden in power system responses, which are subject to the ambient excitation. With the wide applications of wide area measurement system, analysis of low-frequency oscillations on the basis of ambient signals has a good prospect. Based on the characteristic analysis of the response data, a method for oscillation parameters identification of power system on the basis of fast stochastic subspace identification is proposed. FSSI algorithm adopts LQ resolution instead of traditional singular value decomposition process. It not only improves the rate of analysis and calculation, but also gets good recognition accuracy for oscillation frequency, damping ratio and modal shape. The high recognition accuracy and fast calculation of the proposed method is verified by the simulation results of four-machine, two-area system and 16-machine, 68-bus system.