Since Prony algorithm has high requirement to the input signal and sensitive to the noise of analysis data， this paper proposes a fuzzy filtering and Prony algorithm combined online identification method for power system low-frequency oscillation. In the proposed method， the wide area measurement signals are the inputs of a fuzzy filter which can cancel the noise rapidly by simple fuzzy logic， and then the low frequency oscillation modes are obtained online by analyzing the digital signal with the improved Prony algorithm. Simulations on the active power oscillation of branch 302245 of Central China Power Grid（CCPG） and comparisons of the method proposed with the traditional method demonstrate that the proposed method can provide relatively precise analysis results.