We report three studies that examine the accuracy and usefulness of food records among dieting obese patients. In study 1 errors in quantity and calorie estimates for 10 common foods averaged 63.9% (quantity) and 53.4% (calories). The results of study 2 indicate that only 53% of entries in daily food records were specified enough to permit objective estimates of the calories consumed. In study 3, blind raters could not predict weight loss based on subjects' self-recorded behavior changes. Collectively, these results question the utility of food records for estimating energy intake or predicting weight loss.