Aiming at the problem of extracting fault characteristic from gearbox with compound faults under the condition of changing rotating speed, a compound fault diagnosis method for gearbox based on chirplet path pursuit and ensemble empirical mode decomposition (EEMD) is proposed. In the proposed method, the rotating speed is estimated from original vibration signal by chirplet path pursuit algorithm. In order to transform the time domain signal to the angular domain signal, the even angle resampling is carried out to the original signal according to the obtained rotating speed. The angular domain resampling signal is decomposed by EEMD, and the suitable intrinsic mode function (IMF) is selected according to the correlation coefficient. The selected IMF is analyzed by the Hilbert envelope spectrum and the compound fault diagnosis of gearbox is executed according to the envelope spectrum. The results, obtained by the analysis of algorithm simulation and application examples to the gearbox with compound faults consisted of gear local fault and bearing local fault under the condition of changing rotating speed, show that the proposed method can effectively extract fault characteristic from gearbox with compound faults.