As part of the trend towards personalised medicine, surgeons are increasingly using 3D printed replicas for preoperative planning. This raises the question of how reliable these models are. This paper examines the repeatability of manufacturing human mandibles. Five polyamide replicas were produced using selective laser sintering and digitised using structured light scanning. Quantitative comparisons were made using Mimics Software. The differences were analysed graphically, using histograms and kernel density estimates. The mean differences ranged between +0.0274 (SD 0.0671) mm and −0.0284 (SD 0.0629) mm. The median of absolute differences was 0.0308 mm, i.e. 50% of absolute differences were smaller than 31 μm. For the 22,811,168 differences measured, all were between +1.9836 and−2.0526 mm. The proportion of absolute differences below 0.10 mm was between 82.09 and 98.84%, and between 94.43 and 99.90% when using a threshold of 0.20 mm. 99.95% of the absolute differences were below 1.00 mm. In conclusion: 3D printed models may not be identical, even when based on the same imaging study and patient; on the other hand, identical replicas can be obtained with a constant production chain; we recommend that four distinguishing criteria should be used in future investigations: qualitative and quantitative accuracy, repeatability and reproducibility.