Food avoidance is central to patients with anorexia nervosa-restrictive type (AN-R). Competing accounts in experimental psychopathology research suggest that food avoidance may result from automatic, habitual responses or from elevated inhibitory control abilities. This study investigated behavioral trajectories of food avoidance in a novel virtual reality stopping task. Sixty patients with AN-R and 29 healthy controls with normal weight were investigated using a novel, kinematic task in virtual reality. We recorded spatial displacement in stop- and go-trials to virtual food and control objects. Inhibitory control abilities were operationalized by the VR task in stopping performance (i.e., interrupted movement in stop-trials), whereas we also measured habitual avoidance of virtual food across both go- and stop-trials (i.e., delayed movement relative to nonfood objects). In patients with AN-R, hand displacements were shorter to food versus nonfood across stop- and go-trials, reflected in a Stimulus × Group interaction. Healthy controls showed no differences. Importantly, the food-specific effect in AN-R was identical across stop- and go-trials, indicating habitual food avoidance. Moreover, stop error rates (i.e., stop-trials with response) were lower in patients with AN-R. The findings suggest food-specific habitual avoidance and heightened generalized inhibitory control in AN-R. The continuously delayed displacements during active hand movements across stop- and go-trials indicated the persistence of patients' avoidance of food. Experimental research investigates the mechanisms underlying mental disorders such as anorexia nervosa. In this study, we measured interrupted hand movements in response to food pictures or neutral pictures (shoes) in patients with anorexia nervosa and healthy controls. A virtual reality scenario was used. Findings indicated that patients were slower at approaching food, interrupted or not. Key mechanisms of food avoidance can be translated into habit-based treatment options in future research.