Abstract: The first intrinsic mode function(IMF) is generally a multi-component as the result of the abundant frequency component in gear fault signal.Pseudocomponents in low frequency may be produced in empirical mode decomposition(EMD).An improved Hilbert-Huang transform(HHT) method was proposed to overcome these two problems.Frequency span of fault information has been achieved in frequency spectrum, so corresponding binary wavelet coefficient of narrow-band signal could be extracted.The method of correlation coefficient screening was used to remove pseudocomponent in EMD of wavelet coefficient.Local instant energy was used to extract gear fault feature.It is demonstrated that improved HHT method could diagnose the gear fault effectively.