Weight estimation is of great importance in assessing fetal development, yet unavailable in routine fetal MRI. The aim of this study was to develop an automatic fetal body segmentation method and to create a large dataset of volumetric body measurements of normal fetuses. Automatic fetal body segmentation was performed on data obtained from two clinical sites, three MRI systems and two sequences. Using a neural network trained for each sequence, high performance was achieved for both of them. A database of normal fetal volumes with a wide range of gestational age was created and was consistent with ultrasound growth chart.