A method was investigated for measuring the volumes of human livers in vivo from magnetic resonance images and subsequently displaying these livers in three dimensions. Volumetric image sets of phantoms, healthy volunteers, and patients with cirrhotic livers were processed. Two image-processing approaches were compared for accuracy of liver measurements, intrasubject and interobserver variation, and speed of processing. Results indicated that both processing methods had a high degree of volume-measuring accuracy (within 8%), the interobserver measurements had a high coefficient of correlation (r = .9994), the intrasubject measurements had a low coefficient of variation (1.8%), and one method was four to five times faster than the other. The faster and easier of the two image-processing approaches provided satisfactory results for measuring liver volumes, but the slower approach provided more realistic-looking three-dimensional images of the liver.