In view of the similarity between transient components extraction from mechanical vibration signals and the significant speech information detection from noisy environment,a gear crack fault feature extraction method based on auditory attention mechanism is proposed. The saliency map calculation model is introduced and improved combined with the characteristic of fault information. By frequency band division and processing,multiscale Gauss filtering,saliency calculation and integration,the transient impact components in fault signals are finally represented by auditory saliency maps. The effectiveness of the proposed method is verified by experiments with simulation and measured signals of gear crack fault.