The ice pack covering northern seas is composed of a mixture of thick ridged and rafted ice, undeformed ice, and open water. Ice motions determined from satellite remote sensing data show that deformation of the pack takes place along the boundary of large floes. Eulerian continuum sea ice models can simulate this behavior to a degree by capturing the localization of gridded ice strength in some average sense. However, using a discontinuous Lagrangian approach that explicitly models ice floes and the interactions between them, it is possible to simulate both the fracture process that creates floe boundaries and the continued deformation along those floe boundaries. We have developed a granular model of the central Arctic ice pack that consists of thousands of individual grains that can freeze together, fracture apart, and interact through ridging. Accelerations produced by passing weather patterns and sustained quasi-steady deformation cause the model pack to fracture apart into floes composed of one or more grains. When the ice pack is nonuniformly accelerating due to passage of a weather pattern, simulations show that the factors that influence the size of the floes are the tensile strength of the joints between grains, the gradient of the wind field, and the average size of the individual grains. During sustained deformation the pack continues to deform along existing floe boundaries while stresses build and further fracture takes place. In quiet areas of the basin, fractures refreeze. To explore the fracture process during sustained deformation, we run 24-hour basin scale simulations at resolutions from 2.8 km to 14 km. At the end of each simulation we construct a distribution of floe areas. The cumulative distribution of floes at the large end of the distribution is approximately the same at each resolution. As we increase the resolution from 14 km to 2.8 km, the damage zones between large floes become more localized. A log-log plot of the cumulative floe size distributions obtained from the simulations appears linear over several orders of magnitude.