Résumé : Objective: To evaluate the performance of a simple semi-automated method for estimation of fetal weight (EFW) using magnetic resonance imaging (MRI) as compared with two-dimensional (2D) ultrasound (US) for the prediction of large-for-dates neonates. Methods: Data of two groups of women with singleton pregnancy between March 2011 and May 2016 were retrieved from our database and evaluated retrospectively: the first group included women who underwent US-EFW and MRI-EFW within 48 h before delivery and the second group included women who had these evaluations between 35 + 0 weeks and 37 + 6 weeks of gestation, more than 48 h before delivery. US-EFW was based on Hadlock et al. and MRI-EFW on the formula described by Baker et al. For MRI-EFW, planimetric measurement of the fetal body volume (FBV) was performed using a semi-automated method and the time required for measurement was noted. Outcome measure was the performance of MRI-EFW vs US-EFW in the prediction of large-for-dates neonates, both ≤ 48 h and > 48 h before delivery. Receiver–operating characteristics (ROC) curves for each method were compared using the DeLong method. Results: Of the 270 women included in the first group, 48 (17.8%) newborns had birth weight ≥ 90th centile and 30 (11.1%) ≥ 95th centile. The second group included 83 women, and nine (10.8%) newborns had birth weight ≥ 95th centile. Median time needed for FBV planimetric measurements in all 353 fetuses was 3.5 (range, 1.5–5.5) min. The area under the ROC curve (AUC) for prediction of large-for-dates neonates by prenatal MRI performed within 48 h before delivery was significantly higher than that by US (for birth weight ≥ 90th centile, difference between AUCs = 0.085, standard error (SE) = 0.020, P < 0.001; for birth weight ≥ 95th centile, difference between AUCs = 0.036, SE = 0.014, P = 0.01). Similarly, MRI-EFW was better than US-EFW in predicting birth weight ≥ 95th centile when both examinations were performed > 48 h prior to delivery (difference between AUCs = 0.077, SE = 0.039, P = 0.045). Conclusion: MRI planimetry using our purpose-designed semi-automated method is not time-consuming. The predictive performance of MRI-EFW performed immediately prior to or remote from delivery is significantly better than that of US-EFW for the prediction of large-for-dates neonates. Copyright © 2017 ISUOG. Published by John Wiley & Sons Ltd.