This study evaluated the accuracy of body mass index (BMI) based on self-reported weight and height for predicting adult nutritional status. In a cross-sectional study of 3,934 adults (> 20 years) in Pelotas, Rio Grande do Sul, Brazil, a sub-sample of 140 individuals was drawn and weight and height were measured. From the comparison between "measured" and "reported" BMI, the average reported BMI error was estimated and the associated factors were identified. Regardless of nutritional status, women underestimated their "reported" BMI, while in men this information was accurate. Among women, age and income were associated with underestimated BMI in a multivariate analysis. Thus, women over 50 and with lower income underestimated BMI by more than 2 kg/m2. The use of "reported" BMI to predict adult nutritional status can underestimate prevalence of obesity and overestimate that of overweight in women. Correction minimizes this kind of bias, thereby making the data more accurate.