To automatically detect the tool wear state in metal cutting process, an automatic detection method based on image processing was proposed by analysis of the gray levels of a tool wear image in the background area, wear area and the unworn area. Firstly, the automatic determination algorithms of the upper-and lower-thresholds were respectively presented according to the Otsu method and B-spline curve fitting method. Based on the algorithms, the gray-contrasts between wear area and background area as well as wear area and unworn areas were enhanced. By analysis of the stable region within three areas and the non-stable region at two edges in the tool wear image, the local gray-level variance threshold algorithm was presented for the boundary extraction. An adaptive threshold of the local variance was defined, by which the tool wear region was segregated from the tool wear image clearly. On this basis, the morphological method was used to fill out the holes of the segregated part, so that the corresponding geometric parameters of the wear areas were precisely achieved. Experimental results show that the detection errors of the wear width and the wear length are respectively 1.024% and 1.325% when the magnification of 3D microscope is 50, whereas they are 0.661% and 0.995% when the magnification is 100. The proposed method is characterized by strong anti-interference, and higher detection accuracy. It can supply the technology support for improving the tool utilization, guaranteeing the machining quality, and saving the manufacturing cost.