To solve the problem that the vibration signals of a hydraulic pump usually appear with nonlinear and a low signal to noise ratio, we devised a degradation feature extraction method based on improved multi-fractal detrended fluctuation analysis (MF-DFA). First, we proposed an improved MF-DFA based on slip window technology to overcome the problem of data partitioning, then used the MF-DFA to calculate four kinds of multi-fractal parameters of a hydraulic pump. We analyzed four parameters and consequently selected the singular index a0 and the width of multi-fractal spectrum △a as the degradation feature. Finally, the results from analyzing actual data showed that the proposed method effectively recognized the degradation status of the hydraulic pump.