Modelling fracture of individual stones in DEM simulations is of importance in many applications. One important example is mineral processing where it is of interest to fracture the pieces as energy efficiently as possible. Another important case is road and railway structures where fracture of the stones should be avoided to ensure structural integrity. I both these applications, a statistical model of the fracture force is needed mainly due to the variability of the stone shapes. Another complexity is to handle particles of different sizes as brittle materials are known to show a statistical size dependency, i. e that smaller stones are for their size stronger than larger stones. The usual method of modelling the fracture force is to use a Weibull distribution, either with or without a weakest-link description. Using a plain Weibull description, the statistical size dependency is neglected while using a traditional Weibull weakest-link model all variations in fracture force are attributed to the statistical size effect. The aim of this work is to present a statistical fracture force model accounting for variability in shape as well as the statistical size effect. The new model is calibrated against single stone crushing tests for different types of material potentially used for road construction and it is seen that only the new model can accurately describe the statistical outcomes of all four materials. Furthermore, confined compression tests are performed both experimentally and using DEM and the amount of fractured stones are most accurately predicted using the new statistical model for different materials and stone size distributions.