Implicit approach associations are well documented for substances such as alcohol, tobacco, and illicit drugs. This study reports two experiments designed to establish and modify such associations specifically in the food craving domain. Experiment 1 used a pictorial implicit association task to examine approach-avoidance associations with chocolate cues in a sample of 48 undergraduate women. Participants were faster to respond to trials that paired chocolate pictures and approach words, and trials that paired pictures of highly desired food items not containing chocolate and avoid words, than the converse pairings. The magnitude of this approach bias was positively correlated with participants' reported chocolate craving. Experiment 2 examined whether chocolate-related approach associations can be modified. Using a modified implicit association task, 96 undergraduate women were trained to associate chocolate pictures either with approach or with avoid words. As predicted, chocolate-approach associations increased in the approach group and decreased in the avoid group. Additionally, the approach group reported stronger chocolate cravings after training; in contrast, cravings tended to decrease in the avoid group. These results are consistent with incentive- and cognitive-motivational accounts of craving, and support and extend reports of approach biases (and the retraining of those biases) for other substances, including alcohol and cigarettes, to the food domain. They also offer potential scope for curbing unwanted food cravings in the context of problem eating behavior. (Contains 3 tables and 3 figures.)