Cognitive neuroscience, especially in the fields of learning and decision-making, is witnessing the blossoming of computational model-based analyses. Several methodological and review papers have indicated how and why candidate models should be compared by trading off their ability to predict the data as a function of their complexity. However, the importance of simulating candidate models has been so far largely overlooked, which entails several drawbacks and leads to invalid conclusions. Here we argue that the analysis of model simulations is often necessary to support the specific claims about behavioral function that most of model-based studies make. We defend this argument both informally by providing a large-scale (N>300) review of recent studies, and formally by showing how model simulations are necessary to interpret model comparison results. Finally, we propose guidelines for future work, which combine model comparison and simulation.