Rapid prototyping (RP) is a relatively new technology that produces physical models by selectively solidifying UV-sensitive liquid resin using a laser beam. The technology has gained a great amount of attention, particularly in oral and maxillofacial surgery. An important issue in RP applications in this field is how to obtain RP models of the required accuracy. We investigated errors generated during the production of medical RP models, and identified the factors that caused dimensional errors in each production phase. The errors were mainly due to the volume-averaging effect, threshold value, and difficulty in the exact replication of landmark locations. We made 16 linear measurements on a dry skull, a replicated three-dimensional (3-D) visual (STL) model, and an RP model. The results showed that the absolute mean deviation between the original dry skull and the RP model over the 16 linear measurements was 0.62 +/- 0.35 mm (0.56 +/- 0.39%), which is smaller than values reported in previous studies. A major emphasis is placed on the dumb-bell effect. Classifying measurements as internal and external measurements, we observed that the effect of an inadequate threshold value differs with the type of measurement.