The　gear　is　the　key　component　of　rotating　machinery,　so　a　fault　in　the　gear　will　directly　affect　the　condition　of　the　whole　machine＇s　operation.　It　was　difficult　to　extract　the　fault　feature　information　effectively　from　the　vibration　signals　of　a　faulty　gear.　In　the　field　of　fault　diagnosis,　envelope　demodulation　was　one　of　the　most　common　signal　processing　methods.　However,　a　filtering　process　was　required　before　envelope　demodulation.　The　parameters　of　a　filter　were　determined　by　experience,　and　that　has　a　great　influence　on　the　results　of　signal　processing.　The　discrete　wavelet　packet　transform　has　a　larger　energy　leakage　of　frequency　band,　which　obviously　affected　the　results　of　the　envelope　demodulation.　It　is　necessary　to　have　a　method　with　a　lower　energy　leakage　of　the　frequency　band　before　envelope　demodulation.　The　dual　tree　complex　wavelet　packet　transform　(DT-CWPT)　was　a　new　signal　processing　method　that　had　many　good　qualities.　Because　the　energy　leakage　of　the　frequency　band　was　smaller　when　the　signal　was　decomposed　by　a　dual　tree　complex　wavelet　packet　transform,　the　dual　tree　complex　wavelet　packet　transform　was　used　to　extract　the　fault　feature　information　in　the　field　of　fault　diagnosis.　In　this　paper,　first,　according　to　the　characteristics　of　Gaussian　white　noise,　whose　frequency　was　full　of　the　whole　frequency　band,　Gaussian　white　noise　was　decomposed　by　a　dual-tree　complex　wavelet　packet　transform,　and　the　parts　with　energy　leakage　were　regarded　as　a　theoretical　part　band　beyond　the　range　of　the　frequency　components.　Then　the　lower　energy　leakage　characteristic　of　dual　tree　complex　wavelet　packet　transform　was　verified　by　a　quantitative　analysis　method　of　frequency　band　energy　leakage.　A　dual　tree　complex　wavelet　packet　transform　has　an　advantage　in　the　pretreatment　of　envelope　demodulation　compared　with　the　method　of　discrete　wavelet　packet　transform.　Secondly,　the　signal　was　decomposed　layer-by-layer　by　a　dual　tree　complex　wavelet　packet　transform,　and　the　kurtogram　based　on　a　dual　tree　complex　wavelet　packet　transform　could　be　obtained　by　computing　the　spectral　kurtosis　of　every　layer＇s　components.　According　to　the　standard　of　maximum　kurtosis,　the　layer　of　decomposition　and　the　component　about　the　signal　can　be　chosen　automatically　and　accurately.　The　best　layer　of　the　dual　tree　complex　wavelet　packet　decomposition　was　the　layer　of　the　maximum　kurtosis　and　the　component　which　had　the　maximum　kurtosis　was　the　best　component　of　decomposition.　Finally,　the　vibration　signal　of　the　engineering　was　processed　by　the　method　of　spectral　kurtosis　based　on　a　dual　tree　complex　wavelet　packet　transform,　the　best　decomposition　layer　and　component　could　be　chosen,　and　the　fault　feature　information　was　extracted　effectively　by　a　Hilbert　envelope　demodulation,　where　the　feasibility　and　effectiveness　of　the　method　were　verified.　The　research　will　provide　a　reference　for　extracting　the　fault　feature　information　of　a　gearbox　fault　diagnosis　in　rotating　machinery.