A dynamic lung tumor phantom was used to investigate the geometric reconstruction accuracy of a commercial fourâdimensional computed tomography (4DâCT) system. A ball filled with resin, embedded in a cork cube, was placed on a moving platform. Various realistic anteroâposterior (AP) motions were programmed to reproduce the respiratory motion of a lung tumor. Several threeâdimensional (3D) CT and 4DâCT images of this moving object were acquired and compared using different acquisition parameters. Apparent volume and diameter of the ball were measured and compared to the real values. The position of two points (the AP limits of the ball) during the motion in the coordinate system of the CT scanner were also compared with the expected values. Volume error was shown to increase with object speed. However, although the volume error was associated with intraslice artifacts, it did not reflect large interslice inconstancies in object position and should not be used as an indicator of image accuracy. The 3DâCT gave a random position of the tumor along the phantom excursion; accuracy in the assessment of position by 4DâCT ranged from 0.4 mm to 2.6 mm during extreme phases of breathing. We used average projection (AVE) and maximum intensity projection (MIP) algorithms available on the commercial software to create internal target volumes (ITVs) by merging gross tumor volume (GTV) images at various respiratory phases. The ITVs were compared to a theoretical value computed from the programmed ball excursion. The ITVs created from the MIP algorithm were closer to the theoretical value (within 12%) than were those created from the AVE algorithm (within 40%).PACS numbers: 87.53.Xd, 87.56.Mp, 87.57.Ce, 87.59.Fm