The　energy　distribution　of　each　frequency　sub-band　of　acoustic　emission(AE)　signal　after　decomposition　is　related　to　the　type　of　roller　bearing　defects,　AE　signals　measured　from　a　roller　bearing　test　rig　were　decomposed　into　a　number　of　frequency　sub-bands　by　using　harmonic　wavelet　packet,　and　energy　features　associated　with　each　sub-band　were　selected.　The　energy　features　were　then　used　as　inputs　to　a　back-propagation　neural　network　classifiers　for　identifying　the　bearing＇s　fault.　In　the　bearing　fault　recognition,　harmonic　wavelet　packet　was　compared　with　Daubechies　wavelet　packet.　The　experimental　results　indicated　that　the　proposed　fault　diagnosis　method　is　effective　and　can　be　used　for　roller　bearing　fault　recognition.