Complex　nonlinear　and　nonstationary　signals　can　be　adaptively　analyzed　by　the　Hilbert-Huang　transform　through　empirical　mode　decomposition　and　the　Hilbert　transform　to　generate　the　instantaneous　energy.　The　instantaneous　energy　was　able　to　display　the　local　characteristics　of　the　signals　and　had　good　time-frequency　analysis　capability,　it　is　therefore　widely　applied　to　the　analysis　of　vibration　signals　in　the　field　of　gear　fault　diagnosis.　However,　only　a　few　extracted　intrinsic　mode　functions　through　empirical　mode　decomposition　can　reflect　fault　feature　or　closely　related　to　the　faults　but　others　are　irrelevant.　Therefore,　the　fault　feature　of　the　instantaneous　energy　for　all　intrinsic　mode　functions　was　not　obvious　and　the　accuracy　of　diagnosis　was　low.　Aimed　at　solving　this　problem,　a　fault　leading　rate　evaluation　algorithm　was　proposed　that　can　select　those　intrinsic　mode　functions,　which　reflect　fault　features　(it　was　called　the　dominant　intrinsic　mode　function)　from　all　intrinsic　mode　functions.　In　the　paper,　this　algorithm　was　applied　to　gear　fault　feature　extraction.　By　calculating　the　instantaneous　energy　of　the　dominant　intrinsic　mode　function　the　method　could　accurately　extract　gear　fault　feature　and　improve　the　accuracy　of　diagnosis.　Both　simulated　signals　and　experimental　signals　of　a　Klingelnberg　bevel　gear　were　analyzed　to　verify　the　effectiveness　and　correctness　of　the　algorithm.