The relapse prevention model has recently been applied to the treatment of obesity in order to explain why people stop practising caloric restriction and hence stop losing weight. This study attempted to identify environmental, affective and physiological variables characterizing 'high risk' situations that lead individuals to either overeat or eat unplanned meals, thereby violating the rules of dietary abstinence. Thirty-five individuals (5 male; 30 female) at least 20 percent above ideal body weight were instructed to keep a continuous record of their eating behavior during a 10-week behavioral weight loss program. Each eating episode was entered into the computer as a separate behavioral observation resulting in 3066 usable observations. Log-linear analysis of multidimensional contingency tables was used to fit Markov chains to the data in order to examine whether dietary slips (overeating; unplanned eating) could be predicted from the following antecedents or pairs of antecedents: meal, people, place, mood, and hunger. Antecedents found to be associated with conditional probabilities 25 percent greater than the unconditional probabilities for unplanned meals and overeating (very much; some) are reported. Three basic categories of high-risk situations emerge: positive social interactions, negative emotions, and physiological craving. Individual differences were observed in the unconditional probabilities of the target behaviors, in exposure to high-risk situations, and in impulsive eating in response to meal, people, place, and hunger.