In an attempt to improve attention bias modification (ABM), we tested whether an attentional training protocol which featured monetary operant conditioning of eye-gaze to avoid alcohol stimuli in alcohol-dependent patients could reduce attention, craving and relapse to alcohol. We employed a pilot randomized control trial (RCT) with 21 detoxified alcohol dependent patients (48.9 ± 10 years of age, 9 male) from an inpatient and outpatient treatment centre. The novel concealed operant conditioning paradigm provided monetary reinforcements or punishments respective to eye-gaze patterns towards neutral or towards alcohol stimuli along with an 80% probability of a to-be-detected probe appearing following neutral stimuli (ET-ABM group). Patients in the control-group received random monetary feedback and a 50/50 ABM contingency. We compared AB on trained and untrained stimuli and addiction severity measures of obsessive thoughts and desires to alcohol following training. We further assessed addiction severity and relapse outcome at a 3-month follow-up. Results indicate that this attentional retraining only worked for the trained stimuli and did not generalize to untrained stimuli or to addiction severity measures or relapse outcome. Potential explanations for lack of generalization include the low sample size and imbalances on important prognostic variables between the active-group and control-group. We discuss progress and challenges for further research on cognitive training using gaze-contingent feedback.