The need to understand and predict the long- and short term behavior of railway ballast calls for extensive experimental programs as well as the development of reliable numerical models. From the modeling point of view, the granular nature of ballast in connection with the high onset speed of dynamic loading makes it ideal for the application of the Discrete Element Method (DEM). The paper describes the employment of the DEM to simulate ballast behavior in large-scale oedometric testing. Ballast grains are represented by convex polyhedral particles whose shapes are randomly generated via Voronoi tessellation. The effect of the aspect ratio of particles is investigated. A novel algorithm to compute repulsive contact force based on the intersecting volume of polyhedrons is presented. Crushing of grains is included via splitting the particles into smaller polyhedrons when a certain stress-based criterion is fulfilled. The results obtained from the model are compared with the published experimental investigations. (C) 2014 Elsevier B.V. All rights reserved.