Mastering the dynamic characteristics of the pressure pulsations is necessary to ensure high efficiency and security, and extend the life span of the pump turbine (PT) and the pumped storage system. In this research work, the experimental investigation was conducted with the aid of the high-precision equipment to measure the pressure pulsation signals from a PT in the turbine direction. The variational mode decomposition (VMD) and Hilbert marginal spectrum (HMS) are adopted to improve the extraction capabilities of dynamic characteristics in the signals. Through carrying out the VMD, the pressure pulsation signals can be decomposed into several modes. By calculating the average instantaneous frequencies, energy ratios, and correlation coefficients of these modes, the dominant modes are determined, and the trends of the values of root mean square of them with load at typical monitoring points can reflect the dynamic transition of the internal flow status. To be specific, the entire load range can be segmented into three load areas to analyze different physical sources of pressure pulsations. In the low load area, they are mainly induced by the unstable flows of blade passing frequency (BPF) in the vaneless area (VA) and the low-frequency vortex strips in the draft tube (DT). In the intermediate load area, they are mainly caused by the low-frequency vortex strips in the DT. In the high load area, they are mainly caused by the phenomenon of rotor-stator interaction in the VA. By conducting the HMS based on VMD, it can be found that the head changes will affect the amplitudes of dominant frequencies of pressure pulsation signals. Based on the current signal processing results, the high-sensitivity indexes can be used to characterize the generation and dynamic propagation of pressure pulsations.