The rolling bearing of gearbox of electric multiple-unit (EMU) is in a variable speed condition with high temperature and heavy load during operation, which is easy to induce faults such as cracks and pitting corrosion that are difficult to be detected. In order to diagnose the faults of rolling bearing in gearbox of EMU in time and ensure the safe operation of EMU, a rolling bearing fault diagnosis method under variable speed condition was proposed. First, a fusion time-frequency analysis algorithm was proposed, combining the characteristics of no interference term of short-time Fourier transform (STFT) and high time-frequency resolution of Wigner-Ville distribution (WVD), which can improve the time-frequency matrix accuracy of variable speed signal analysis. Then, the dynamic path planning method was improved considering the limitation that this method cannot deal with the normalized time-frequency matrix, and the speed curves in the fused time-frequency matrix were extracted. Furthermore, an order analysis method of interpolation resampling was proposed. The interpolation resampling of the original signal was performed according to the speed. The signal was reconstructed in the angular domain, and the corresponding order spectrum was obtained to realize the fault diagnosis of rolling bearing. Finally, the proposed fault rolling bearing diagnosis method was verified on test bench, and results showed that the proposed method could effectively extract the variable speed curves of the rolling bearing when the speed of the EMU changed, and accurately identify the fault types of the rolling bearing in the gearbox.