Cognitive biases for alcohol-related stimuli have been shown to play a major role in the maintenance of alcohol addiction, thus making their study an issue of paramount importance. Such cognitive biases have been traditionally investigated with low-level, non-inferential processes, such as the Stroop task [for a review see Psychol. Bull. 120 (1996) 3]. However, explanations for these effects [for example, the current-concerns explanation; Motivation, Attention, and Volition (1987) 336] posit that biases should mediate cognitive processes more generally, not limited to a single aspect of cognition. In the present study, we addressed this issue within an experimental paradigm extensively used for the study of learning and inference. Participants who were heavy users of alcohol were impaired in the task when the stimuli were alcohol-related but not impaired when the stimuli were neutral. By contrast, participants who were light users of alcohol performed equivalently on the alcohol-related and neutral versions of the learning task. (C) 2002 Elsevier Science Ireland Ltd. All rights reserved.