The invention discloses a rolling bearing degradation index extraction method based on time-frequency feature separation. The method comprises the steps of firstly calculating a time-frequency spectrum of a bearing original signal, secondly, determining the number of maximum value points at the same moment and the number of maximum value points at the same frequency in the time-frequency matrix, determining the harmonic distribution position and range in combination with the line accumulation result of the time-frequency matrix, retaining a harmonic distribution area, filtering other areas, and performing inverse short-time Fourier transform on the processed time-frequency matrix to a time domain so as to realize extraction of harmonic components, after subtracting the harmonic component from the original signal, extracting the fault impact component through threshold noise reduction, and finally calculating the relative root-mean-square value of the extracted fault impact component to serve as a rolling bearing degradation index. According to the method, the prominent harmonic components in the time-frequency spectrum are filtered in advance based on the time-frequency characteristics so as to enhance the fault impact components, serious distortion generated when the impact components are extracted through threshold noise reduction is avoided, the rolling bearing degradation index is calculated for the extracted fault impact components, and the index can effectively represent the bearing degradation process.