Aims To examine whether alcohol-related attentional bias (AB) can be reduced by training heavy drinkers to attend to soft drinks as an alternative to alcohol. Diminishing AB is important because AB has been suggested to be a significant factor in the development, maintenance and relapse of addictive behaviours. AB was trained in a clinically relevant design, and we studied the generalization of this training. Design, participants and intervention We assigned randomly 106 heavy drinking male college and university students to the attentional re-training (AR; modified visual-probe task) or control condition (standard visual-probe task). Setting Laboratory at Maastricht University. Measurements We measured the effects of AR on the visual-probe task with stimuli that were presented in the AR and with new stimuli, and on an alternative measure of AB, the flicker paradigm. We further measured effects on craving and preference for either an alcohol beverage or a soft drink. Findings After AR, participants had learned to avoid alcohol stimuli and had developed an AB for soft drinks. This effect was restricted to stimuli used in the AR. The flicker task, where AB for alcohol was found in both the AR and control groups, was not affected by the AR. No effect was found on craving and the preference task. Conclusions Although heavy drinkers can learn to attend selectively to an alternative category for alcohol, a single AR is not sufficient to decrease symptoms of problem drinking.