Aims The main aim of this study was to test whether automatic action-tendencies to approach alcohol can be modified, and whether this affects drinking behaviour. Design and participants Forty-two hazardous drinkers were assigned randomly to a condition in which they were implicitly trained to avoid or to approach alcohol, using a training variety of the alcohol Approach Avoidance Test (AAT). Participants pushed or pulled a joystick in response to picture-format (landscape or portrait). The pictures depicted alcoholic or non-alcoholic drinks. Participants in the avoid-alcohol condition pushed most alcoholic and pulled most non-alcoholic drinks. For participants in the approach-alcohol condition these contingencies were reversed. After the implicit training, participants performed a taste test, including beers and soft drinks. Automatic action tendencies at post-test were assessed with the AAT, including both trained and untrained pictures, and with a different test (Implicit Association Test, IAT). We further tested effects on subjective craving. Results Action tendencies for alcohol changed in accordance with training condition, with the largest effects in the clinically relevant avoid-alcohol condition. These effects occurred outside subjective awareness and generalized to new pictures in the AAT and to an entirely different test using words, rather than pictures (IAT). In relatively heavy drinking participants who demonstrated changed action tendencies in accordance with their training condition, effects were found on drinking behaviour, with participants in the approach-alcohol condition drinking more alcohol than participants in the avoid-alcohol condition. No effect was found on subjective craving. Conclusions Retraining automatic processes may help to regain control over addictive impulses, which points to new treatment possibilities.