Due　to　the　influence　of　the　strong　background　noise,　the　harmonic　feature　of　the　shock　fault　on　the　rolling　bearing　can　be　identified　in　the　frequency　field　when　the　fault　reaches　some　severe　level.　Therefore,　the　incipient　fault　feature　can　not　be　extracted　effectively　through　frequency　spectrum　analysis.　By　using　the　band　pass　filtering　feature　of　wavelet　transforms,　the　signal　can　be　decomposed　to　specific　frequency　band.　Furthermore,　the　reconstruction　of　single-level　decomposed　signal　can　separate　the　valuable　components　from　the　noise　successfully.　The　same　results　can　be　obtained　through　global　reconstruction　after　threshold　denoising　on　the　high　frequency　decomposition　coefficients.　The　further　processing　results　of　the　denoised　signal　on　the　rolling　mill　have　showed　that　the　low　frequency　information　containing　fault　feature　is　extracted,　and　the　characteristic　frequency　can　be　captured　from　the　frequency　spectrum　of　denoised　signal.　As　a　result,　the　incipient　fault　can　be　positioned　precisely.