The aim of this prospective study was to evaluate the usefulness of fetal weight estimation in a prospective study, based on formulae derived from a previous retrospective study, using forward stepwise multiple regression analysis. Three-dimensional ultrasound was used to make reproducible measurement of limb circumferences, which formed the basis of our models. A total of 213 women with singleton pregnancies were scanned in the week prior to delivery and the following variables were measured: biparietal diameter, head circumference, transverse abdominal diameter, abdominal circumference, femur length, thigh circumference and arm circumference. The whole population was divided into three subgroups according to the abdominal circumference (< 10th centile, 10-90th centile, and > 90th centile). We demonstrated the need for only two models for accurate fetal weight estimation, one for the small-for-dates fetuses and a second for the others. The most accurate results were obtained for the macrosomic fetuses with a standard deviation of 8.8%. Our data confirm the usefulness of measurement of fetal thigh circumference for the the small-for-dates-fetuses and arm circumference for the other groups. We concluded that the use of three-dimensional ultrasound could facilitate the accurate prediction of fetal weight.