Objective: The purpose of this study was to evaluate the errors of three-dimensional mandibular surfaces generated using magnetic resonance imaging (MRI) when computed tomography (CT) was set as the gold standard. Settings and sample population: Seven patients with orthognathic deformities who had undergone CT and MRI scans were included in the study. Materials and Methods: Mandibular surfaces were generated on each CT and MR image by the surface-rendering method. Intra-individual reliability between CT and MRI was statistically tested by the confidence limits of agreement (LOA) for systematic error, 95% confidence interval minimal detectable change (MDC₉₅) for random error and intra-class correlation coefficient (ICC). Results: The average total error was 1.6 mm. The greatest MDC₉₅ was observed in the coronoid region in all directions. The other regions showed MDC₉₅ values of < 1.8 mm (transvers direction), 3.5 mm (vertical direction) and 1.7 mm (antero-posterior direction). ICCs showed ‘almost perfect’ agreement with respect to all regions. Conclusion: Random errors were quantified for 3-D rendering of the mandible from MRI data. Although the coronoid region showed the greatest errors, the other regions of the mandibular surfaces generated using MRI were able to be evaluated.