Fast　kurtogram　(FK)　can　be　applied　to　rolling　bearing　and　gearbox　fault　diagnosis.　A　finite　impulse　response　filter　is　constructed　to　divide　the　signal　into　two　or　three　components　containing　different　frequency　information　from　spectrum.　Then,　whether　the　spectral　kurtosis　of　each　component　has　failed　is　determined.　The　method　is　fast,　but　sometimes　it　is　unable　to　accommodate　those　bands　that　actually　contain　fault　information.　It　may　even　result　in　the　inability　to　detect　significant　fault　information　from　the　extracted　components.　Therefore,　FK　has　obvious　and　even　fatal　flaws　in　dividing　the　frequency　domain.　A　new　method　to　divide　boundaries　from　the　spectrum　to　optimize　the　FK　is　proposed.　It　is　named　empirical　fast　kurtogram　(EFK).　Firstly,　the　components　representing　the　spectrum　trend　in　the　Fourier　transform　function　of　the　signal　spectrum　are　extracted　and　the　minimum　point　sequence　is　searched.　Taking　the　position　of　the　minimum　value　in　the　spectrum　as　the　boundary　sequence,　the　Meyer　wavelet　is　used　to　construct　the　filter　and　reconstruct　the　signal　components　to　obtain　the　kurtosis.　Finally,　a　new　empirical　fast　kurtogram　is　constructed　and　the　fault　information　is　extracted　from　the　frequency　band　with　the　largest　kurtosis.　The　method　divides　boundaries　according　to　the　spectrum　trend,　which　can　effectively　avoid　the　irrational　phenomenon　caused　by　the　average　spectrum　division　in　the　FK　method.　The　effectiveness　of　the　method　is　demonstrated　by　the　analog　signal　and　the　inner　and　outer　ring　fault　signals　of　the　rolling　bearing.　©　2020,　Nanjing　Univ.　of　Aeronautics　an　Astronautics.　All　right　reserved.