INTRODUCTION: Computer navigation plays an increasingly important role in craniomaxillofacial surgery. The difficulties in computer navigation at the craniomaxillofacial site lie in the accurate transmission of the dataset to the operating room. This study investigates the accuracy of the dental-splint registration method for the skull, midface, and mandible. MATERIAL AND METHODS: A synthetic human skull model was prepared with landmarks and scanned with cone beam computer tomography (CBCT). Two registration splints fixed the mandible against the viscerocranium in two different positions (closed vs. open). The target registration error was computed in all 278 landmarks spread over the entire skull and mandible in 10 repeated measurements using the VectorVision(2) (BrainLAB Inc., Feldkirchen, Germany) navigation system. RESULTS: If registered in the closed position an average precision of 2.07mm with a standard deviation (SD) of 0.78mm was computed for all landmarks distributed over the whole skull. Registration in the open position resulted in an average precision of 1.53mm (SD=0.55mm). For single landmarks the precision decreases linearly with distance from the reference markers. The longer the three-dimensional distance between the registration points, the more precise the computer navigation is, mainly in the most posterior area of the cranium. CONCLUSION: Our findings in the cranium are comparable with those of other studies. Artificial fixation of the lower jaw via splint seems to introduce no additional error. The registration points should be as far apart from each other as possible during navigation with the splint.