Although some studies have analysed the accuracy of self-reported weight and height data in survey studies, no attention has been paid to the accuracy of the body mass index (BMI, kg/m2), derived from these data and often considered as a reliable indirect estimate of relative body weight. Based on data from about 3400 participants in a study, questionnaire self-reports of these anthropometric data were compared to measured weight and height. A flat slope syndrome (under-reporting of high values, over-reporting of low ones) was found for weight and BMI for both sexes. A low sensitivity of BMI based on self-reported values were found, especially for the obese sub-population. Reliance upon questionnaire-derived self-reports will lead to considerable underestimation of the prevalence of obesity. Only 55 per cent of obese women and 60 per cent of obese men according to measured values were correctly classified as such. Correction of self-reported data according to regression models describing the relationship between questionnaire self-reports and medical records of body size parameters should be considered for some types of epidemiologic studies such as those using BMI as a categorized variable. The statistical rationale for the models used here is discussed.