AbstractOver the last decades, the assessment of alcohol cue‐reactivity gained popularity in addiction research, and efforts were undertaken to establish neural biomarkers. This attempt however depends on the reliability of cue‐induced brain activation. Thus, we assessed test–retest reliability of alcohol cue‐reactivity and its implications for imaging studies in addiction. We investigated test–retest reliability of alcohol cue‐induced brain activation in 144 alcohol‐dependent patients over 2 weeks. We computed established reliability estimates, such as intraclass correlation (ICC), Dice and Jaccard coefficients, for the three contrast conditions of interest: ‘alcohol’, ‘neutral’ and the ‘alcohol versus neutral’ difference contrast. We also investigated how test–retest reliability of the different contrasts affected the capacity to establishing associations with clinical data and determining effect size estimates. Whereas brain activation, indexed by the constituting contrast conditions ‘alcohol’ and ‘neutral’ separately, displayed overall moderate (ICC > 0.4) to good (ICC > 0.75) test–retest reliability in areas of the mesocorticolimbic system, the difference contrast ‘alcohol versus neutral’ showed poor overall reliability (ICC < 0.40), which was related to the intercorrelation between the constituting conditions. Data simulations and analyses of craving data confirmed that the low reliability of the difference contrast substantially limited the capacity to establish associations with clinical data and precisely estimate effect sizes. Future research on alcohol cue‐reactivity should be cautioned by the low reliability of the common ‘alcohol versus neutral’ difference contrast. We propose that this limitation can be overcome by using the constituent task conditions as an individual difference measure, when intending to longitudinally monitor brain responses.