The vibration signal of start-up or slowdown process can be regarded as a dynamic response to a wide frequency range excitation, which contains more information than steady state vibration. So it deserves more attention to extracting fault feature by rotor start-up or slowdown process’ analysis. Aiming at the defect that conventional start-up or slowdown process’ analysis applying STFT to get amplitude and phase will lose transient information, a method based on Vold-Kalman filtering order tracking (VKF-OT) and holospectrum is proposed. By this method, amplitude and phase can be obtained directly from the complex envelop of each order component, so it can preserve the transient information well and get rid of the average effect of STFT. In this paper, amplitudes and phases of the vibration from two perpendicular directions of a rotor section are obtained and holo-watefall curve is plotted according to holospectrum theory, then fault feature extraction and transient balancing are fulfilled. The experiment results show that the method can extract the typical fault feature of the rotor and the transient balancing works well.