The accuracy in determining the three-dimensional anatomy of a vessel network by computed tomography (CT) is evaluated using a glass model of a pulmonary artery. The dynamic spatial reconstructor (DSR), a high temporal resolution, volumetric, roentgenographic, CT scanner, was used to scan the model. The glass of the model had a roentgen attenuation coefficient mu = 0.55 cm-1, which is approximately equivalent to the 20% dilution of contrast medium to be expected in the pulmonary arterial tree following a contrast agent bolus injection of 2 ml/kg in the right atrium. The model was scanned inside a 20 cm diameter Plexiglas cylinder with a 1 cm thick wall (mu congruent to 0.2 cm-1) to simulate the chest wall of a 20 kg dog, and it was filled with potato flakes to simulate lung parenchyma (mu congruent to 0.06 cm-1). In one 0.011 s scan, information for reconstruction of a stack of images of transaxial sections was recorded. Sequential scans were performed to obtain data for either maximum transaxial resolution (14 angles of view every 0.0167 s, 120 parallel slices each 1.8 mm thick) or maximum axial resolution (eight angles of view every 0.0167 s, 240 parallel slices each 0.9 mm thick) reconstructions. Estimated detectable "vessel" size, cross-sectional area, branching angle, and interbranch segment length were determined as a function of imaged slice thickness, orientation of section image, and number of angles of view (i.e., scan duration) used to make images. Retrospective selection of 0.05 s duration scan apertures at sequential 0.5 s intervals was used to simulate a typical, retrospectively gated reconstruction from a DSR scan. Using these reconstructed images, 2 mm diameter "vessels" could be readily detected and their structure quantitated. Comparing direct measurements and DSR estimates, cross-sectional area (SEE = 3 mm2), branching angles (SEE = 2 degrees), and segment length (SEE = 1 mm) all had a correlation coefficient greater than 0.99, and the regression lines showed no significant differences from the lines of identity (p greater than 0.05).