The　vibration　signals　of　gearbox　incipient　failure　often　contain　strong　noise,　which　results　difficulty　in　fault　feature　extraction　by　the　conventional　denoising　method,　such　as　threshold　based　method.　Thus,　a　new　method　based　on　dual-tree　complex　wavelet　transform　(DT-CWT)　and　morphological　component　analysis　(MCA)　was　proposed.　In　the　processing,　the　signal　was　firstly　processed　by　DT-CWT　to　gain　the　coefficients　of　different　layers.　Secondly,　MCA　was　employed　to　denoise　the　coefficient　which　was　more　periodic.　Then,　the　denoised　signal　with　weak　fault　feature　could　be　gotten　from　a　following　single　reconstruction.　Finally,　the　fault　characteristic　frequency　could　be　located　accurately　by　simple　envelope　spectrum　analysis.　A　simulate　signal　and　incipient　failure　vibration　signal　of　mill　gearbox　were　processed　using　this　method,　and　the　results　show　that　the　method　can　remove　the　strong　background　noise　in　the　signal　effectively,　and　has　better　effect　than　single　MCA　and　soft　threshold　method,　and　get　a　more　clear　fault　characteristic　frequency,　thereby　providing　a　new　method　for　gearbox　incipient　fault　diagnosis.　©　2016,　Editorial　Department　Journal　of　Aerospace　Power.　All　right　reserved.