The use of three-dimensional (3D) models of the dentition obtained from cone beam computed tomography (CBCT) is becoming increasingly more popular in dentistry. A recent trend is to replace the traditional dental casts with digital CBCT models for diagnosis, treatment planning and simulation. The accuracy of these models was previously assessed through comparing linear physical and radiographical measurements. However, this assessment technique is both observer and landmark dependent. The accuracy of 3D CBCT teeth reconstructions is yet to be reliably measured. To assess the accuracy of 3D CBCT reconstructions of the teeth using a semi-automated and observer-independent method and to assess the influence of field of view (FoV) selection on reconstruction accuracy. Fully dentate upper and lower dry human jaws, placed in a plastic box and immersed in water, were scanned using CBCT with small, medium and large FoV. The teeth were then scanned separately using MicroCT. Cone beam computed tomography and MicroCT 3D teeth models were compared, and mean surface difference was calculated per tooth for each FoV. Mean and (maximum) differences between MicroCT and CBCT were 120 +/- 40 (max. 679) microm, 157 +/- 39 (max. 824) micro and 207 +/- 80 (max. 862) microm for the small, medium and large FoV, respectively. Cone beam computed tomography models were larger than MicroCT because of larger voxel size. Our results indicate that CBCT may provide accurate 3D reconstructions of the teeth that can be useful for some clinical applications.