Abstract: A frequency-time based pattern recognition method was presented for recognizing different disturbance modes in fiber distributed disturbance sensing system based on M-Z interferometer by using the frequency of output interferometer signal in relation to the external disturbance signal. The frequency-time characteristic was measured by using the rate which the output signals crossed the preset average level. Then frequency-time characteristic was segmented and the corresponding feature element model could be set up. The optimum models were selected by using dynamic time warping (DTW) algorithm, and then they were sent to the artificial neural network (ANN) to carry out training and judging. This method could effectively reduce the difficulty of training and judging the signal and time sensitivity of the ANN, and improve the adaptability for the environment. Experimental results illustrate that this method can effectively distinguish different disturbance events such as short-term, long-term, radial and irregular event. The average recognition speed is less than 0.26 s and average accuracy is great than 97%.