Ore rocks are commonly identified as heterogeneous materials, that is, their various micro-structures are characterized by inherent natural randomness. Reasonably, heterogeneity is an important factor that controls damage initiation, fracture propagation, and in turn final breakage failure or fragmentation process within raw rocks. This paper focuses on the internal spatial variability in terms of mineral distribution, and the objective is to numerically investigate the effects of different randomness feature of fine scale structure, as described by variable magnitudes of spatial correlation length parameter, on the rolling compression induced breakage failure of individual rock specimen on a horizontal table. A simple and direct algorithm was developed to generate specimens characterized by random fields of prescribed spatial correlation, which are obtained by a weighted average of random fields without spatial correlation. The finite element method with conversion to the smoothed particle hydrodynamics method was adopted to simulate the progressive deformation, fracture and final breakage failure by roller compression. From parametric study results, it was found that the spatial correlation length parameter value can significantly influence the breakage failure and the corresponding size reduction efficiency for ore rock prisms. Specifically, the progressive fragmentation patterns, and the characteristic size of main broken pieces show obvious length value-related distribution features. The numerical study given in this paper can definitely help understanding the pulverizing mechanism of various polycrystalline particles in roller mills and the optimization of the grinding works for better efficiency. (C) 2016 Elsevier Ltd. All rights reserved.