Fault feature information extraction of broken tooth is often contaminated out by strong production background noises.Aiming at the problem,Autogram method was applied to extract the feature information.In the method,maximum overlapping discrete wavelet packet transform was utilized to decompose a contaminated signal of broken tooth fault into several signals,and each signal was called “node”.In order to depict the feature information comprehensively,unbiased autocorrelation of squared envelope for each node was computed,then one node was selected as the data source to be analyzed.For the sake of effective feature information extraction,threshold processing was introduced,and spectrum analysis could be more comprehensive.The method is effective through comparing the simulation signal sand the test ones.