A　redundant　lifting　wavelet　packet　analysis　based　on　variable　parameter　was　presented,　which　was　used　to　extract　the　weak　fault　features　of　bearing　submerged　in　background　noise.　Through　different　choices　of　parameters　in　the　design　formula　of　predictor　based　on　least　square　method　of　fitting,　six　asymmetric　wavelets　with　various　characteristics　were　constructed,　which　were　then　respectively　used　for　redundant　lifting　wavelet　packet　decomposition　to　signal　layer　by　layer.　All　the　results　obtained　by　decomposition　were　adopted　to　establish　the　objective　function　of　minimum　norm,　through　which　the　optimal　wavelet　that　best　matches　the　feature　information　is　selected　for　each　node　signal.　The　node　signals　got　by　last　decomposition　were　utilized　for　wavelet　packet　energy　analysis,　while　the　node　signal　with　maximum　energy　was　chosen　for　single　branch　reconstruction　and　envelop　spectrum　analysis.　The　proposed　method　above　is　applied　to　process　the　engineering　data　of　bearing　with　fault　and　good　results　are　gained　by　which　the　effectiveness　of　this　method　is　well　validated.