Of the many steps involved in producing high quality three-dimensional (3D) images of CT data, the data acquisition step is of greatest consequence. The principle of "garbage in, garbage out" applies to 3D imaging--bad scanning technique produces equally bad 3D images. We present a formal study of the effect of two basic scanning parameters, slice thickness and slice spacing, on image quality. Three standard test objects were studied using variable CT scanning parameters. The objects chosen were a bone phantom, a cadaver femur with a simulated 5 mm fracture gap, and a cadaver femur with a simulated 1 mm fracture gap. Each object was scanned at three collimations: 8, 4, and 2 mm. For each collimation, four sets of scans were performed using four slice intervals: 8, 4, 3, and 2 mm. The bone phantom was scanned in two positions: oriented perpendicular to the scanning plane and oriented 45 degrees from the scanning plane. Three-dimensional images of the resulting 48 sets of data were produced using volumetric rendering. Blind review of the resultant 48 data sets was performed by three reviewers rating five factors for each image. The images resulting from scans with thin collimation and small table increments proved to rate the highest in all areas. The data obtained using 2 mm slice intervals proved to rate the highest in perceived image quality. Three millimeter slice spacing with 4 mm collimation, which clinically provides a good compromise between image quality and acquisition time and dose, also produced good perceived image quality. The studies with 8 mm slice intervals provided the least detail and introduced the worst inaccuracies and artifacts and were not suitable for clinical use. Statistical analysis demonstrated that slice interval (i.e., table incrementation) was of primary importance and slice collimation was of secondary, although significant, importance in determining perceived 3D image quality.