To solve the problem of ending effect of empirical mode decomposition ( EMD) in HilbertHuang transform, a method to extend the ends of data based on auto regressive integrated moving average model (ARIMA) is presented in this paper . To take advantage of ARIMA in modeling the nonstationary data, the original signal is extended on both ends before EMD, then after HHT, the additive parts are removed and the remaining is the processed result. It is proved by both a simu lation signal and a rotor misalignment experiment that the improved algorithm ca n effectively suppress the ending effect, and the fault features of signal are clearer. Moreover, the parameters of ARIMA are easy to determine, so that the pro posed algorithm can be used conveniently.