Failure of level ice against an inclined marine structure has been simulated using a two-dimensional finite-discrete element method. The simulation model is deterministic, but very sensitive to initial conditions. This allowed us to create ice load data and to study the evolution of the ice failure process. The mean load, the standard deviation, and the maximum load increased during the ice failure process. Ice thickness had a strong effect on the ice load and also the plastic limit of ice had an effect, especially when the ice was thick. The coefficient of variation of the ice load was initially high and then continuously decreased during the ice failure process. This suggests that the ice failure process did not reach a stationary stage, showed high probability of extreme peak loads, and that distributions with a constant coefficient of variation should not be used for this kind of ice loading processes. Extreme value analysis, with the assumption that the ice load follow a log-normal distribution on each time step, fitted the data well and suggested that the maximum ice load increases with sample size. Further, it appears that the peak ice loads are bounded by the crushing capacity of the ice even in the case of an inclined structure. The statistical analysis of simulated ice load data further suggests that observations from short interaction processes, or with small sample sizes, may lead to very inaccurate ice load estimates.