Computational psychiatry provides novel approaches and tools using formal models of brain function to investigatepsychiatricdisorders 1 andtranslatefindingsfromanimal research to human clinical settings. 2 In this Viewpoint, we discuss how computational models of human learning can help overcome previous limitations to progress in addiction research. We also discuss their potential for translation and back translation that might provide for a more in-depth examination of human findings within the context of animal experiments.Drug addiction is characterized by habitual or compulsive drug intake in the face of aversive consequences. 1 Many patients describe that cues associated with drug use (conditioned cues) elicit strong drug urges and trigger drug intake in spite of conscious decisions to remain abstinent. 3,4 However, examination of underlying mechanisms is limited because of the subjective character of self-reported drug craving. Computational models of behavior can circumvent such subjective reporting bias, as these models can quantify fine-grained aspects of behavior as well as identify key computational steps in the underlying learning mechanisms (including reward prediction errors) that can then be directly associated with brain signatures. 1 Regarding cue-induced drug craving and habitual or compulsive drug intake, 2 basic learning-associated mechanisms are implicated: Pavlovian mechanisms, including Pavlovian-to-instrumental transfer (PIT), 3 and the relative dominance between goal-directed and habitual behavioral control. 4 Paradigms to assess these mechanisms have been developed in animal experiments and translated to human studies, with emergent findings now ready for back translation into basic research that can include an in-depth molecular characterization.Considering Pavlovian mechanisms, drug-associated cues (eg, paraphernalia associated with drug consumption) evoke a psychological expectation of drug effects, which in turn drive craving and are associated with instrumentaldrug-seekingbehavior.TwotypesofPIT,general and specific, have been identified in animals and humans, with unique neural correlates characterizing each type. 5 IngeneralPIT,cuesgenerallyenhanceactionspaired with several different outcomes. In specific PIT, cues enhance actions associated with an outcome that is specific to the cue. Animal and human studies implicate the centralnucleusoftheamygdalaandnucleusaccumbens(NAc) in general PIT, while the basolateral amygdala, the NAc shell, and ventrolateral putamen are more closely associated with specific PIT. Translation from animal experiments to humans has helped reveal that enhanced activity within NAc is associated with general PIT in prospective relapsers. 5