This　paper　presents　a　new　bearing　fault　diagnosis　approach　using　wavelet　domain　operator-based　signal　separation.　The　measured　vibration　signal　is　first　preprocessed　using　the　continuous　wavelet　transform　(CWT)　to　filter　unwanted　noise.　Then　an　operator-based　signal　separation　approach,　called　null　space　pursuit　(NSP),　is　applied　to　decomposing　the　signal　into　a　series　of　subcomponents　and　residues　in　accordance　with　their　characteristics.　Subsequently,　the　selected　subcomponent　with　the　maximum　Kurtosis　value　is　analyzed　by　the　envelop　spectrum　to　identify　potential　fault-related　characteristic　frequency　components.　Experimental　studies　from　a　real　wind　turbine　gearbox　have　verified　the　effectiveness　of　the　presented　approach　for　bearing　fault　diagnosis.