Abstract: The joints of different scale levels in coal and rock are staggered and the distribution law is very different, which has an important influence on their deformation and failure characteristics. The fracture scale level, which plays an important role in the mechanical behaviour of coal rock, is screened and the discrete fracture network (DFN) model is constructed on the basis of this scale. While ensuring the accuracy of numerical calculation, the influence of redundant fracture scale on model construction and computer processing power can be reduced, which can provide a new way to improve the efficiency of numerical calculation of rock mechanics. Based on CT scanning and image processing technology, the three-dimensional fracture structure and geometric characteristic parameters of coal samples were extracted, and the RFPA^3D-CT was embedded for uniaxial loading numerical analysis and compared with the uniaxial compression test results of the samples. The meso-mechanical parameters of the numerical model of coal samples were determined, and the fracture scale levels, which play an important role in controlling the mechanical behaviour of coal samples, were further investigated step by step. On this basis, the Baecher model and Monte Carlo stochastic simulation theory are used to construct the equivalent DFN model according to the geometric distribution characteristics of the main control fracture, and the validity of the model is verified by numerical simulation. The results show that: The main control fracture scale ( trace length ) of the coal specimen sample is not less than 14 mm. Compared with the model of deleting the cracks above this scale, the failure mode changes from ‘2’ type to oblique ‘1’ type, and the failure peak strength increases by 61.79%. The existence of cracks above this scale plays a decisive role in the numerical calculation results of the specimen model. The equivalent DFN model is constructed based on the main crack scale of the specimen. The geometric properties of the joints are basically consistent with the geometric parameters of the real cracks based on CT scanning, and the average error is 5.16%. The mechanical properties of the equivalent DFN model under uniaxial loading are relatively consistent with the six groups of full-crack models after CT scanning. The average error of the peak strength of the two is 13.41%, and the calculation efficiency of the failure calculation time step is improved by 14.59%.