OBJECTIVE To assess the influence of body build on the bias and limits of agreement for estimates of body fat obtained from anthropometric prediction equations when compared with the same data obtained by dual energy X-ray absorptiometry (DEXA). SURVEY DESIGN AND SUBJECTS: Ninety-one premenopausal women, aged between 20 and 54 years, were chosen to represent a range of skeletal body build (relative sitting height 0.50-0.56) and body fatness [body mass index (BMI) 18-34 kg/m2]. Measurements of weight, sitting height, stature, skinfold thickness, waist, umbilical and hip circumference and total body resistance and reactance were made on all subjects by standard techniques after an overnight fast. A DEXA measurement of total body fat, fat-free soft tissue and total body bone mineral mass was also obtained within 2 weeks of the anthropometric assessment. RESULTS At the group level the mean difference (bias) between DEXA and the anthropometric estimates of body fat was similar for all three anthropometric estimates ranging from 2.7 kg with impedance to 1.8 kg with skinfold thickness. The 95% limits of agreement were also similar, ranging from +/- 5.3 kg with body mass index to +/- 4.1 kg with impedance. Umbilical circumference, BMI and the amount of bone mineral expressed as a proportion of the fat-free soft-tissue mass were all significantly (P < 0.01) correlated with the level of bias between DEXA and the anthropometric estimates of body fat. This was not the case for relative sitting height or measures of body fat distribution. Regression equations which included BMI or umbilical circumference in combination with the predicted estimates of body fat essentially eliminated the association between the level of bias in predicted body fat and the level of body fatness. They also reduced the 95% limits of agreement between DEXA and the anthropometric estimates of body fat. CONCLUSIONS Using DEXA estimates of body fat as the standard of reference our results suggest that the comparability and precision of body fat estimates derived from age- and/or sex-specific anthropometric prediction equations based on skinfolds and BMI, but not impedance, can be improved by adjusting for differences in BMI and umbilical circumference respectively.