Abstract Rationale Severe alcohol use disorder (SAUD) is a psychiatric condition linked to cerebral and cognitive consequences. SAUD is notably characterized by an overactivation of the reflexive/reward system when confronted with alcohol-related cues. Such overreactivity generates a preferential allocation of attentional resources toward these cues, labeled as attentional biases (AB). Theoretical assumptions have been made regarding the characteristics of AB and their underlying processes. While often considered as granted, these assumptions remain to be experimentally validated. Aims We first identify the theoretical assumptions made by previous studies exploring the nature and role of AB. We then discuss the current evidence available to establish their validity. We finally propose research avenues to experimentally test them. Methods Capitalizing on a narrative review of studies exploring AB in SAUD, the current limits of the behavioral measures used for their evaluation are highlighted as well as the benefits derived from the use of eye-tracking measures to obtain a deeper understanding of their underlying processes. We describe the issues related to the theoretical proposals on AB and propose research avenues to test them. Four experimental axes are proposed, respectively, related to the determination of (a) the genuine nature of the mechanisms underlying AB; (b) their stability over the disease course; (c) their specificity to alcohol-related stimuli and (d) their reflexive or controlled nature. Conclusions This in-depth exploration of the available knowledge related to AB in SAUD, and of its key limitations, highlights the theoretical and clinical interest of our innovative experimental perspectives capitalizing on eye-tracking measures.