Abstract: The threshold segmentation of CT image of coal is the prerequisite for the reconstruction of 3D real structure. A porosity calculating method was proposed based on grey level distribution of CT images obtained under uniaxial compression. Combined with the gray scale histogram, CT images were experienced three-value-segmentation of fracture, coal matrix and inclusion by inverse process from experimental porosity. The three-value-segmentation results obtained by Otsu method, maximum entropy method and the improved gray histogram method show that maximum entropy method can not get a good segmentation between coal inclusion and coal matrix, and Otsu method can not effectively split crack and coal matrix for its low sensitivity to dramatic variation for gray value in small area of the image. However, the improved gray histogram method eliminates the disadvantages of the two methods mentioned above. The threshold of coal matrix and inclusion is obtained with higher reliability and accuracy based on the experimental results, and a better three-value-segmentation of image is obtained by the improved grey histogram method.