We compared the accuracies of adiposity distribution measures (waist and hip circumferences, subscapular and triceps skinfold thicknesses, waist-hip and subscapular-triceps ratios) with a measure of adiposity extent (body mass index) for screening non-insulin dependent diabetes mellitus (NIDDM). Of 521 subjects (218 men and 303 women) who had 2 h oral glucose tolerance tests, 43 men and 28 women were found to have NIDDM. Allowing for age, ethnicity, and family history in analyses of covariance, the subscapular skinfold thickness and the ratio of the subscapular to triceps skinfold thicknesses were significantly higher (P < 0.05) in both the men and women found to have diabetes than in their non-diabetic counterparts. The waist-hip ratio was also significantly higher in diabetic women; however, the elevation was not as marked in diabetic men (P = 0.06). The body mass index did not differ between diabetic and non-diabetic men and was only moderately elevated in diabetic women (P = 0.04). Receiver-operator curves were employed to examine the relative accuracies of the body mass index, waist-hip ratio, and subscapular skinfold thickness for screening NIDDM. In both men and women, the waist-hip ratio and subscapular thickness were superior to the body mass index. These data suggest that certain measures of adiposity distribution are more accurate than measures of overall adiposity extent for screening NIDDM and that they may be useful in screening programmes.