Rolling　bearings　are　one　of　the　most　important　components　in　rotating　machinery.　It　is　important　to　accurately　determine　the　center　frequency　and　bandwidth　of　the　resonant　frequency　band　for　bearing　fault　diagnosis.　There　are　two　problems　with　the　existing　methods　for　extracting　bearing　fault　characteristics.　First,　due　to　the　unreasonable　spectrum　segmentation,　the　determined　resonant　frequency　band　contains　only　partial　fault　information　or　hidden　irrelevant　information.　Finally,　because　of　the　interference　of　the　accidental　impact,　the　correct　fault　characteristic　information　cannot　be　extracted.　To　solve　the　above　problems,　a　time-frequency　domain　scanning　empirical　spectral　negentropy　method　(T-FSESNE)　based　on　spectral　negentropy　(NE)　and　empirical　wavelet　transform　(EWT)　is　proposed　in　this　paper.　The　signal　is　filtered　twice　by　EWT　filter:　Firstly,　the　central　frequencies　of　all　resonance　side　bands　are　determined　by　using　frequency-domain　spectral　negentropy,　and　then　the　optimal　bandwidth　of　the　resonance　side　bands　is　determined　by　using　time-domain　spectral　negentropy.　According　to　the　determined　center　frequency　and　bandwidth,　each　component　is　extracted　and　analyzed　by　envelope　spectrum　to　realize　bearing　fault　diagnosis.　The　validity　of　the　extracted　methods　is　verified　by　bearing　fault　simulation　and　experimental　signals.　The　results　show　that　not　only　the　interference　of　accidental　impact　can　be　effectively　avoided　but　also　the　optimal　center　frequency　and　bandwidth　can　be　determined　quickly　and　accurately.　More　importantly,　this　method　can　determine　the　position　of　multiple　resonance　sidebands,　which　is　more　suitable　for　the　analysis　of　complex　fault　vibration　signals　in　rolling　bearings.