A discrete element model (DEM) is used to study the mechanical behaviour of a pre-fractured brittle medium subjected to triaxial loadings where progressive failure can occur. The initial discrete fracture network (DFN) is based on a fractal distribution and it is used to characterize the influence of clustering and size distribution of pre-existing fractures, on the strength of fractured rock masses. The proposed approach captures the progressive failure process occurring in the rock matrix as shearing displacement takes place along pre-existing discontinuities. The results show that the mechanical behaviour of fractured rock masses is mainly dependent on the fracture intensity. However, for a given fracture intensity, the strength can exhibit a 50 per cent variability depending on the size distribution of the pre-existing fractures. This difference can be attributed to the different mechanisms that involve sliding and crushing of blocks in the case of large interconnected fractures or progressive failure of the rock matrix through coalescence of cracks in the case of small unconnected fractures. Clustering of fractures was found to influence the spatial variability of the mechanical properties and therefore to have a scale effect on strength. The results outline the relevance of three parameters, the power-law exponent of the fracture size distribution, the clustering fractal dimension which fixes the fracture-to-fracture correlation number and the fracture intensity, to characterize the mechanical behaviour of rock masses.