Accumulating evidence shows that overweight individuals exhibit an attentional bias for food and that this bias can be modified. This study investigated the longevity of such modification effects. Using a dot probe paradigm, a community sample of overweight and obese women (N = 104) was trained to direct attention towards (‘attend’) or away from (‘avoid’) food pictures. Participants completed five weekly training sessions. Attentional bias was measured before and after training, at 24 h and one-week follow-up. To increase generalisability, at each of the post-training and follow-up assessments, participants were shown a mix of old and new food pictures. They also completed another implicit bias measure, i.e. a word stem task. Attentional bias for food increased in the ‘attend’ group and decreased in the ‘avoid’ group. These retraining effects were maintained at 24 h and one-week follow-up, and extended to new food pictures. Participants in the ‘avoid’ group also produced relatively fewer food words on the word stem task than those in the ‘attend’ group. Results are consistent with predictions of cognitive-motivational models that attentional biases are malleable. They further suggest that attentional bias modification, which targets the implicit processes that underlie the heightened food responsivity in overweight individuals, could help combat pathological (over)eating.