Computed tomography (CT) based three-dimensional (3-D) lifelike models have proved to be of great value, especially in craniomaxillofacial surgery. They improve and facilitate diagnosis, therapeutic planning, model operations, and definitive treatment in tumor surgery, traumatology, dysgnathia, alveolar atrophy, and congenital and asymmetrical malformations. From 1988 to 1998, 760 stereolithography (STL) and milled 3-D models were employed in our department. These two production methods have become the msot commonly used approaches, and the question as to which is preferable is the subject of controversy in the literature. Using two test models the STL and the milling method were compared with regard to production method and accuracy, and the resulting differences in indications for their use are discussed. The limiting factor for increased accuracy is the distance between each pair of CT scans. Milled models show a higher precision compared to STL models within the CT-scan plane, but the more the measurements deviate from this plane (becoming oblique) the more inaccurate the data becomes. STL exhibits greater deviations, but the inaccuracy is randomly distributed. The mean variation from the original was 0.81 mm for STL and 0.54 mm for milled models. Although 90% of the milled and 80% of the STL values are within a deviation of +/- 1 mm, it should be noted that maximum deviations of 3.15 mm in milled models and 2 mm in STL have been shown. Both methods are sufficiently accurate for clinical use. In standard cases involving the upper and lower jaw, malar bone, orbita, and calvaria the shorter production time and lower costs make milled models preferable. In special cases, in which hollows and fine structures play a major role (e.g., those involving the skull base, paranasal sinuses, inner ear, and mandibular canal), STL is indicated.