Alcohol use disorder (AUD) is highly prevalent in Veterans and is associated with poor treatment outcomes. Evidence-based treatments (EBTs), such as Cognitive Behavioral Therapy (CBT), show utility in reducing maladaptive drinking and restoring healthy goals, but more than two thirds of treatment completers relapse within a year. Anhedonia, i.e., a reduced interest in rewarding activities, is commonly observed in AUD patients and predicts poorer response to psychiatric treatment. Anhedonic symptoms manifest through a dopaminergic deficit state in the frontostriatal reward circuitry, which contributes to the reinforcement of alcohol use as a fast-acting mood-regulating strategy. Moreover, while anhedonia is known to hinder goal-directed reward behavior, this presentation is multi-faceted, and may impact different reward processing mechanisms, e.g., learning, anticipating, pursuing, and receiving rewards. However, the degree to which these reward behavior mechanisms relate to AUD outcomes remains poorly understood and have not been investigated in Veterans. Understanding these relationships could help shed light on factors that facilitate or hinder AUD recovery and may help identify specific neurobehavioral targets not adequately addressed in standard care to guide the development of new interventions. To address these questions, we propose to use computational modeling and functional neuroimaging to mechanistically characterize the relationships between anhedonia and drinking behavior in treatment-seeking Veterans with AUD. Computational modeling, particularly in concert with neuroimaging, provides detailed mechanistic insights into the neurobiology of cognitive processes, and such methods have been shown to be more accurate in predicting clinical outcomes, relative to standard behavioral and neuroimaging analysis. We will capitalize on this approach to delineate robust anhedonia-driven predictors of AUD treatment responsiveness in Veterans. A total 124 Veterans with AUD will be recruited through a recently funded clinical trial as they enroll in 12 weeks of group-based CBT (standard care EBT). Participants will further be randomized to an adjunctive computer-based protocol, i.e., control (sham) or approach-avoidance training (AAT). Because AAT is aimed at reducing patients’ habitual approach bias towards alcohol cues, it may provide incremental benefits in shifting approach behaviors from reflexive to more goal-driven. At baseline, 12-, and 24-weeks after treatment onset, participants will complete a clinical assessment and 2 multi-arm bandit (MAB) tasks (in classic and social conditions, to be compared in exploratory analyses), in which they must choose on each trial from among a set of options with unknown reward probabilities, with the goal of maximizing total rewards. Concurrent brain activity will be measured in a subset of 62 participants. Using Bayesian modeling, we will derive individual parameters of reward learning (i.e., perceived sta...