Abstract: A tool wear fault diagnosis method based on local mean decomposition (LMD) was proposed to effectively monitor the cutting tool condition. By using LMD, acoustic emission signals would be adaptively decomposed into a series of product functions (PF). Average energy of each PF was extracted from the first 8 product functions containing main fault information, which was represented as the feature vector. The cutter wear characteristics can be recognized according to the variance contribution of the parameters of feature vector, which could be obtained from normal, medium and severe wear states, respectively. Experimental results show that the average energy of PF increases with the increment of tool wear. Especially, the average energy in high frequency markedly increases. Test results also have verified the effectiveness of the presented method.