Randomized trials of medications for alcohol dependence (AD) often report no differences between active medications. Few studies in AD have tested hypotheses regarding which medication will work best for which patients (ie, precision medicine). The PREDICT study tested acamprosate and naltrexone vs placebo in 426 randomly assigned AD patients in a 3-month treatment. PREDICT proposed individuals whose drinking was driven by positive reinforcement (ie, reward drinkers) would have a better treatment response to naltrexone, whereas individuals whose drinking was driven by negative reinforcement (ie, relief drinkers) would have a better treatment response to acamprosate. The goal of the current analysis was to test this precision medicine hypothesis of the PREDICT study via analyses of subgroups. Results indicated that four phenotypes could be derived using the Inventory of Drinking Situations, a 30-item self-report questionnaire. These were high reward/high relief, high reward/low relief, low reward/high relief, and low reward/low relief phenotypes. Construct validation analyses provided strong support for the validity of these phenotypes. The subgroup of individuals who were predominantly reward drinkers and received naltrexone vs placebo had an 83% reduction in the likelihood of any heavy drinking (large effect size). Cutoff analyses were done for clinical applicability: individuals are reward drinkers and respond to naltrexone if their reward score was higher than their relief score AND their reward score was between 12 and 31. Using naltrexone with individuals who are predominantly reward drinkers produces significantly higher effect sizes than prescribing the medication to a more heterogeneous sample.