Résumé : Low insight is reported as a risk factor for relapse among patients treated for alcohol use disorders. However, to date, little is known on why patients with low insight are at higher risk for relapse. In this study, we tested the hypothesis that an implicit preference for alcohol over abstinence predicts relapse in patients with low, but not high, alcohol insight. Participants consisted of 77 patients who had received treatment for severe alcohol use disorder in a hospital in France. During hospitalization, they completed a self-report measure of insight and an implicit association test to assess implicit preference for alcohol over abstinence. The primary outcome was relapse assessed one month after discharge. Control variables were gender, age, cognitive deficit, anxiety, depression, craving, and impulsivity. Data were analysed using logistic regression analysis. After adjusting for demographic and clinical variables, relapse was predicted by the interaction between insight and implicit preference for alcohol but not by their main effects alone. Implicit preference for alcohol predicted relapse among patients with relatively low insight, but not among those with relatively high insight. These findings suggest that patients with low insight and strong implicit preference for alcohol are at a higher risk of relapse. Clinicians may therefore focus on and tailor specific interventions to prevent relapse in this vulnerable and at-risk population.