Many factors influence the measurement uncertainty when using computed tomography for dimensional metrology applications. One of the most critical steps is the surface extraction phase. An incorrect determination of the surface may significantly increase the measurement uncertainty. This paper presents an edge detection method for the surface extraction based on a 3D Canny algorithm with sub-voxel resolution. The advantages of this method are shown in comparison with the most commonly used technique nowadays, i.e. the local threshold definition. Both methods are applied to reference standards and industrial parts and the comparison of the uncertainties obtained by both methods is presented.