Approach bias modification (ApBM) has shown promise in addiction treatment, but effects are small and ecological validity suffers from completing trainings in the laboratory. Providing app-based trainings via mobile phones could increase training efficacy. One-hundred-and-thirty-one smokers seeking abstinence attended a smoking cessation intervention and were thereafter randomized to one of three conditions: (a) app-based ApBM-training; (b) app-based sham training; (c) no training. App trainings were performed on 14 consecutive days at home. Behavioral and self-report data was assessed at pretest, posttest, and a 6-week follow-up. App-based ApBM led to stronger reductions in cigarette smoking as compared to no training. However, both training variants led to stronger declines in nicotine dependence and the sham training was superior in reducing alcohol consumption. Although approach biases for positive cues increased following ApBM training, this effect did not mediate treatment outcome. Other smoking-related cognitive biases did not change after training. Our results can inform future research in the optimization and advancement of ApBM treatment for addiction.