As　gear　fault　vibration　signal　is　always　nonlinear　and　nonstationary　and　always　with　a　strong　background　noise　which　result　in　difficulty　of　fault　feature　extraction,　a　new　method　based　on　dual-tree　complex　wavelet　transform　and　local　projective　method　is　proposed.　As　a　improved　method　of　the　conventional　discrete　wavelet　transform(DWT),　dual-tree　complex　wavelet　transform　has　many　advantages　over　DWT,　such　as　the　improvement　of　frequency　aliasing　and　oscillations　of　wavelet　coefficients　which　is　the　key　to　the　method　proposed.　Local　projective　method　for　nonlinear　time　series　has　a　good　ability　of　signal　period　strengthen　and　noise　suppression,　which　fits　for　wavelet　coefficients　denoising.　Firstly,　the　fault　signal　is　decomposed　by　dual-tree　complex　wavelet　transform　to　obtain　the　coefficients　of　different　layers.　Secondly,　the　nonlinear　time　series　method　is　used　to　strengthen　the　periodicity　of　the　coefficient　whose　amplitude　is　more　periodic,　and　then　do　soft-threshold　denoising.　Finally,　the　fault　characteristic　signal　can　be　obtained　by　coefficient　reconstruction.　The　fault　frequency　can　be　located　accurately　by　Hilbert　envelope　spectrum　analysis.　The　simulation　and　engineering　application　showed　the　effectiveness　of　the　method　in　early　gear　fault　diagnose.　©,　2015,　Nanjing　University　of　Aeronautics　an　Astronautics.　All　right　reserved.