Compound　faults　diagnosis　of　bearings　and　gears　in　gearboxes　is　a　full-challenging　task.　The　rational　usage　of　meshing　resonance　is　probable　to　be　one　of　the　breakthroughs　to　solve　this　problem.　However,　the　existing　detection　method　of　meshing　resonance　is　sensitive　to　noise　and　has　risks　of　false　demodulation.　Meanwhile,　it　is　difficult　to　guarantee　that　isolated　components　only　contain　one　fault　characteristic.　Based　on　the　mentioned　problems,　a　spectrum　of　full-band　preprocessing　and　2-D　separation　method　is　raised.　First,　the　inverted　editing　method　of　the　original　signal　is　proposed　to　reduce　noise　in　the　full　frequency　band.　Second,　the　established　resonance　detection　diagram　and　the　introduced　amplitude-level　decomposition　technique　separate　the　processing　results　in　the　frequency　and　amplitude　dimensions.　Finally,　the　separate　components　are　detected,　respectively,　by　envelope　analysis.　Both　the　simulation　analysis　and　experimental　verification　do　support　the　effectiveness　of　this　method.　In　addition,　this　article　compares　the　existing　method　and　solves　the　essential　problems　as　well.