This study tested the effects of a new cognitive-bias modification (CBM) intervention that targeted an approach bias for alcohol in 214 alcoholic inpatients. Patients were assigned to one of two experimental conditions, in which they were explicitly or implicitly trained to make avoidance movements (pushing a joystick) in response to alcohol pictures, or to one of two control conditions, in which they received no training or sham training. Four brief sessions of experimental CBM preceded regular inpatient treatment. In the experimental conditions only, patients' approach bias changed into an avoidance bias for alcohol. This effect generalized to untrained pictures in the task used in the CBM and to an Implicit Association Test, in which alcohol and soft-drink words were categorized with approach and avoidance words. Patients in the experimental conditions showed better treatment outcomes a year later. These findings indicate that a short intervention can change alcoholics' automatic approach bias for alcohol and may improve treatment outcome.