Objective. To compare the accuracy of fetal weight estimation in large fetuses using four ultrasound formulae. Design. Prospective comparative study. Setting. University Hospital, Lund, Sweden. Population. Large-for-gestational age fetuses (n= 114) at a routine ultrasound examination in the third trimester. Methods. The Persson & Weldner two-dimensional formula was compared with the Hart et al. two-dimensional formula incorporating maternal weight and with the Lee et al. and Lindell & Maršál three-dimensional formulae. The formulae are based on two-dimensional measurements of fetal head, abdomen and femur, and three-dimensional volumetry of fetal abdomen and thigh. Main outcome measures. Accuracy in detection of fetuses with birthweight >4000 and >4500 g. Results. For fetuses >4000 g, the Lindell & Maršál three-dimensional formula showed significantly smaller mean absolute percentage error than the Persson & Weldner two-dimensional and the Lee et al. three-dimensional formulae (p= 0.04 and p < 0.001, respectively). No significant difference between the Lindell & Maršál three-dimensional and the Hart et al. two-dimensional formulae was found. A receiver operating characteristic curve showed a higher detection rate for fetuses with birthweight >4500 g using a three-dimensional compared with a two-dimensional technique. The best performance in detecting fetuses with birthweight >4500 g was given by the Lindell & Maršál three-dimensional formula; for estimated fetal weight >4300 g, the detection rate was 93% and false-positive rate 38%. Conclusions. The ability to detect macrosomic fetuses in a preselected high-risk group was higher using fetal weight estimation with a three-dimensional compared with a two-dimensional ultrasound technique, with or without the inclusion of maternal weight.