Rolling　bearings　are　key　components　of　rotating　machinery　and　easy　to　be　damaged　due　to　complex　working　conditions.　Thus,　it　is　of　great　significance　to　detect　bearing　faults　effectively.　This　paper　proposes　a　new　nonlinear　filtering　technique　named　differential　spectral　amplitude　modulation　(DSAM),　which　can　automatically　extract　different　signal　components　based　on　their　energy　levels　without　any　pre-defined　parameters.　The　core　of　this　algorithm　is　to　adjust　the　energy　distribution　in　a　spectrum　by　using　the　time-domain　differential　of　a　signal,　which　can　not　only　preserve　different　constituents　of　an　original　signal,　but　also　affect　signal　characteristics　by　adaptively　enhancing　the　elements　in　high-frequency　region.　Moreover,　an　indicator　is　proposed　to　optimize　the　outcome,　which　can　weaken　the　interference　of　invalid　components　and　obtain　more　evident　fault　information.　The　rationality　of　this　method　is　verified　by　two　simulated　signals　and　experimental　signals　of　bearing　outer　ring,　inner　ring　and　multiple　faults.　Comparisons　with　other　commonly　used　approaches　demonstrate　the　advantages　of　the　proposed　method.　©　2022　Elsevier　Ltd