Abstract：Rock is a multiscale material，and in the past，the research of rock in macroscopic scale is much more than that in mesoscale. But the damage and failure of rock under loading are caused by the propagation and coalescent of cracks at mesoscale，so it is helpful to understand the mechanical behaviours of rock by quantitatively investigating on mesostructure of rocks. The quantitative test on mesostructure of rock is designed based on damage mechanics. A great deal of mesostructural images of Jinping marble in Sichuan Province have been obtained by means of scanning electron microscopy(SEM). The mesostructural images of marble are processed by regional growing theory based on image processing technique. According to stereology theory，the program is compiled by MATLAB to enhance the details of images and execute the image segmentation. The mesostructural information of rock microcracks such as length，angle，width，area，perimeter，etc. are obtained from binary images after segmentation. Then，the mesostructural information of microcracks is analyzed by statistical theory and the distribution regularities of microcrack parameters are proposed. Based on the statistical regularities of the mesostructural parameters，the representative element volume(REV) of rock in mesoscale is simulated by using the Monte Carlo theory. Finally，the initial damage tensor of REV of rock in mesoscale is obtained according to the geometrical damage mechanics；and it is used in the G. Swoboda damage model to obtain the stress-strain relations under the conditions of uniaxial and triaxial compression loads. By comparing the simulated result with the actual experiment result of stress-strain relationships，it shows that the simulation of two conditions nicely agree with in-situ test results.
