For　the　rolling　bearing　diagnosis,　how　to　identify　the　fault　feature　effectively　is　the　key　issue.　Due　to　the　resonance　modulalion　characteristic　induced　by　shock　fault　of　the　rolling　bearings,　the　wavelet　transform　technology　can　extract　the　modulation　information　effectively.　On　the　other　hand　as　there　are　no　fixed　kernel　functions　in　wavelet　analysis　the　transform　results　are　closely　related　to　the　wavelet　base　function　types.　According　to　shock　and　modulalion　characteristic　of　localized　fault,　how　to　select　the　proper　wavelet　base　function　is　discussed　in　this　article　Through　analyzing　the　simulation　signal　of　outer　race　fault,　the　base　functions　for　the　discrete　wavelet　transform　are　optimized.　The　results　have　shown　that　dmey　wavelet　mother　function　is　prior　to　other　wavelet　functions　in　the　shock　fault　feature　extraction.　Furthermore,　demodulation　technology　based　on　Hilbert　transform　is　used　to　analyze　the　detailed　wavelet　decomposition　coefficient　which　contains　the　modulation　phenomenon.　And　the　fault　feature　can　be　identified　obviously.　Finally,　the　vibration　signal　collected　from　fault　bearing　in　the　wire　rolling　mill　is　decomposed　using　optimized　dmey　wavelet.　The　further　FFT　analysis　on　low　frequency　wavelet　decomposition　coefficient　can　also　identify　the　incipient　fault　feature　successfully.