In dentomaxillofacial radiology, cone-beam computed tomography (CT) is used to give fast and high-resolution 3-dimensional images of bone with a low dose of radiation. However, its use for quantitative measurement of bone density based on absolute values (Hounsfield units, HU) as in multidetector CT is still controversial. We know of no in vivo study of 3-dimensional merging software that will reliably match identical bone areas of cone-beam and multidetector CT datasets. We studied 19 multidetector, and 19 cone-beam, CT scans of the skull. The two datasets were fused, corresponding points were identified for measurement, and we compared mean density. We used linear regression to analyse the relation between the two different scanning methods, and studied a total of 4180 measurements. The mean time interval between scans was 5.2 (4.7) months. Mean R(2) over all measurements was 0.63 (range 0.22 - 0.79) with a mean internal consistency (Cronbach's α) of 0.86 (range 0.61 - 0.93). The strongest linearity, seen at the left mastoid, was R(2)=0.79 with high internal consistency (Cronbach's α 0.89), and the weakest was at the left zygomatic bone with R(2)=0.22 and Cronbach's α=0.61. Measurements of bone density based on cone-beam and multidetector CT scans generated in vivo showed high and reproducible internal consistency but poor linearity.