Introduction : To evaluate the accuracy of CAD/CAM generated splints in orthognathic surgery by comparing planned versus actual post-operative 3D images. Methods : Specific planning software (SimPlant ® OMS Standalone 14.0) was used to perform a 3D virtual Le Fort I osteotomy in 10 fresh human cadaver heads. Stereolithographic splints were then generated and used during the surgical procedure to reposition the maxilla according to the planned position. Pre-operative planned and postoperative 3D CT scan images were fused and imported to dedicated software (MATLAB ®) 7.11.) for calculating the translational and rotational (pitch, roll and yaw) differences between the two 3D images. Geometrical accuracy was estimated using the Root Mean Square Deviations (RMSD) and lower and upper limits of accuracy were computed using the Bland & Altman method, with 95 % confidence intervals around the limits. The accuracy cutoff was set at +/− 2 mm for translational and ≤ 4° for rotational measurements. Results : Overall accuracy between the two 3D images was within the accuracy cutoff for all values except for the antero-posterior positioning of the maxilla (2.17 mm). The translational and rotational differences due to the splint were all within the accuracy cutoff. However, the width of the limits of agreement (range between lower and upper limits) showed that rotational differences could be particularly large. Conclusion : This study demonstrated that maxillary repositioning can be accurately approximated and thus predicted by specific computational planning and CAD/CAM generated splints in orthognathic surgery. Further study should focus on the risk factors for inaccurate prediction.