Using facts about scientists’ incentives in doing research, I argue that a purely heuristic defense of scientific modeling should be cautiously approached. Firstly, due to the need for mathematical tractability, those employing analytic models have an incentive to employ known mathematical structures. This makes it unlikely that their models can suggest novel conclusions. Secondly, while those employing simulations are generally better off in this respect, a look at the details of the way the models are being used is necessary here, too, as there are often prior simulations that can be used as templates for new ones.