Objectives: To examine the effect of measurement error in dietary data on the relationship between diet and body mass index (BMI). To correct for the effect of measurement error on diet-BMI association by using replicate measurements of diet. The effect of measurement error on diet-BMI relationship was simulated, and its implications are discussed. Design: Prospective study design. Setting: The first and second China Health and Nutrition Survey conducted in 1989 and 1991, respectively. Subjects: Three thousand, four hundred and seventy-nine adults age 20-45 y at the 1989 survey. Methods: Statistical methods were used to demonstrate the effect of measurement error in dietary data on the diet-BMI association. Results: By using the average of three replicate 24 h dietary recalls, the attenuation of diet-BMI association was reduced substantially. The regression coefficients of fat and energy intakes differed markedly from those computed by using only single measurement of diet. Conclusions: Measurement error in dietary data may significantly attenuate the diet-disease association. Where appropriate, specific emphasis may be needed to address the problem of measurement error in the study of diet-disease relationship. Spnosorship: This research was supported by the National Institute of Health, the Carolina Population Center and the Nutrition Institute, University of North Carolina at Chapel Hill.