This paper presents a fully automated algorithm for geometry assessment of the mandible. Anatomical landmarks could be reliably detected and distances were statistically evaluated with principal component analysis. The method allows for the first time to generate a mean mandible shape with statistically valid geometrical variations based on a large set of 497 CT-scans of human mandibles. The data may be used in bioengineering for designing novel oral implants, for planning of computer-guided surgery, and for the improvement of biomechanical models, as it is shown that commercially available mandible replicas differ significantly from the mean of the investigated population.