Much of the foundation of clinical practice, psychiatric epidemiology, and research into the etiology, course, prevention, and treatment of alcohol use disorder (AUD) rests on psychiatric diagnosis. However, existing research has failed to adequately exploit empirical techniques and existing databases to derive criteria considered optimal with respect to predicting external correlates. The current project adopts a novel approach to deriving new diagnostic criteria sets and rules for AUD. Utilizing the 2010 (N = 24,120) and 2013 (N = 23,627) National Survey on Drug Use and Health (NSDUH; Substance Abuse and Mental Health Services Administration [SAMHSA], 2011, 2014) data sets, we performed a statistical optimization procedure, using complete enumeration, on participants 21 or older who had consumed at least 1 alcoholic beverage in the past year. The goal was to maximize the distance (based on Cohen's d) between mean levels of the optimization criteria (i.e., consumption and functional impairment) in those with an AUD diagnosis versus those without. In contrast with current convention, AUD is derived transparently using a data-driven approach. The best solution included 9 criteria with a diagnostic threshold of 3, while the second-best solution comprised 5 criteria with a threshold of 2. External validation demonstrated both solutions perform similarly, suggesting it is appropriate to use either, depending on the goal of the diagnosis. Overall, statistical optimization approaches can yield highly efficient criteria sets and rules, although multiple, near equivalently performing solutions can be generated. (PsycINFO Database Record (c) 2019 APA, all rights reserved).