The authors hypothesized that ultrasound-generated measurements of the fetus could be used to predict body composition for children of diabetic mothers. Seven ultrasonically identifiable fetal morphometric parameters were measured on 21 fetuses of diabetic mothers within 7 days of delivery. These seven individual measurements were then matched to neonatal morphometric estimations of body composition by means of stepwise regression analysis. Ultrasound variables which were found to show significant correlation to the newborn estimates of body composition were combined to generate an equation that could predict neonatal lean body mass and fat mass. Neonatal lean body mass could be predicted with R = .93 (P less than .001, 95% confidence limits .80-.98). Neonatal fat mass could be predicted with R = .81 (P less than .001, 95% confidence limits .55-.92). Prospective evaluation of the regression equations in an additional 18 patients showed no significant difference between ultrasound-predicted body composition and newborn morphometric estimation of body composition. The authors conclude that estimations of fetal body composition are possible, and may provide a valuable tool in the identification of fetal growth abnormalities.