The study presented in this thesis aims to numerically explore the micro-mechanisms underlying rock fracture and fragmentation under dynamic loading. The approach adopted is based on the Discrete Element Method (DEM) coupled to the Cohesive Process Zone (CPZ) theory. It assumes rock material as assemblage of irregular-sized deformable fragments joining together at their cohesive boundaries. The simulation, which is referred to as Cohesive Fragment Model (CFM), takes advantage of DEM particle/contact logic to handle the fragments and boundaries in between. In this idealization, mechanical properties of particle and more dominantly those of contact control macroscopic response of the particle assemblage. A rate-dependent orthotropic cohesive law is developed for DEM contacts to capture rock material specific features, e.g. brittleness, anisotropy and rate-dependency. Rock experimental behavior is then modeled in order to assess individually the sensitivity of results to grain size, confining pressure, micromechanical parameters, stored strain energy, loading rate etc. The thesis is organized to approach the problem systematically. First, CFM application for static analysis is examined. It is shown that CFM quantitatively and qualitatively predicts compressive and tensile failure of hard and soft rocks as well as shear strength, dilatancy and degradation of rough rock joints. CFM micro-parameters, i.e., stiffness of particle and strength, stiffness, and friction of contact are calibrated using a combination of statistical disciplines and original closed-form expressions. The calibration process provides useful physical interpretation for each micro-parameter in terms of standard rock mechanical properties. These interpretations enable to understand how macroscopic behavior of rock material originates from its mineral microstructure. Energy needed to fully open a contact, the contact energy numerically represents material fracture energy in CFM. Experimental investigations suggest that fracture energy is independent of loading rate in quasi-static circumstances. Thus, contact energy is simply assumed as constant in static analysis. However, simulation on fast fracturing by CFM warns that this assumption causes serious deviations in fracture dynamic analysis. Laboratory observations reveal that fast-moving fracture consumes more energy than slow-moving one does. This inspires to consider contact energy as variable and rate-dependent to provide the model with the appropriate prediction of the fracture energy release process. Applying this new approach, fracture behavior of PMMA plates is investigated under different levels of stored strain energy. As the final stage, dynamic fracture toughness of rock samples, measured by the split-Hopkinson pressure bar test, is simulated and promising results are obtained. They demonstrate how numerical modeling can practically aid experimental methods in terms of measurement verification, error estimation, and performing appropriate corrections. The studies suggest that DEM is an effective and convenient tool to investigate fracture and fragmentation problems. While predictions by continuum models are restricted only to crack initiation, simulation by DEM made it possible to track both the initiation and progression of fracture over time by following consecutive damage of contacts. Moreover, the research specifically demonstrates that the proposed contact model properly predicts the experimental behavior of rock fracture under static and dynamic loading. This result verifies the model validity and adequacy for rock fracture analysis.