Dixon imaging techniques derive chemical shift-separated water and fat images, enabling the quantification of fat content and forming an alternative to fat suppression. Whole-body Dixon imaging is of interest in studies of obesity and the metabolic syndrome, and possibly in oncology. A three-point Dixon method is proposed where two solutions are found analytically in each voxel. The true solution is identified by a multiseed three-dimensional region-growing scheme with a dynamic path, allowing confident regions to be solved before unconfident regions, such as background noise. 2π-Phase unwrapping is not required. Whole-body datasets (256 × 184 × 252 voxels) were collected from 39 subjects (body mass index 19.8-45.4 kg/m2), in a mean scan time of 5 min 15 sec. Water and fat images were reconstructed offline, using the proposed method and two reference methods. The resulting images were subjectively graded on a four-grade scale by two radiologists, blinded to the method used. The proposed method was found superior to the reference methods. It exclusively received the two highest grades, implying that only mild reconstruction failures were found. The computation time for a whole-body dataset was 1 min 51.5 sec ± 3.0 sec. It was concluded that whole-body water and fat imaging is feasible even for obese subjects, using the proposed method. Magn Reson Med 63:1659–1668, 2010. © 2010 Wiley-Liss, Inc.