The question of the relationship between gesture production and mental models was explored in three experiments focusing on inference making when learning from reading a scientific text. Participants engaged in one-on-one interviews after reading an illustrated tutorial on the human circulatory system. Participants gestured more frequently when responding to mental model-based inference test questions (Experiment 1; N = 22) than textbase or general knowledge questions. Learning from a text lacking illustrations resulted in comparable inference making (Experiment 2; N = 48) but elevated gesture production. Restricting gesture production with hand tapping led to inferior inference making (Experiment 3; N = 79). Results support an embodied cognition view of mental model formation where inference making stimulates gesture production, and restricting gestures selectively impairs inference making. The results are interpreted within the Gesture as Model Enactment (GAME) framework. Activation of motor control systems via gesture support simulated actions that contribute to mental model formation. Implications for learning and assessing knowledge are explored.