This paper presents a new method for extracting the dynamic characteristics of the draft tube of a hydraulic turbine based on EMD（empirical mode decomposition） and AR（auto regressive） model. The signals processed in advance are decomposed with EMD， thus the intrinsic mode functions（IMFs） containing characteristic frequencies can be obtained. The AR model can then be developed for every IMF， and the parameters of AR model can be used as the characteristic parameters for fault mode recognition. As an example， the monitoring signals of the pressure fluctuation of the draft tube are processed with the proposed method. It is shown that the method effective for extracting fault information.