@Helvetosaur The lack of accuracy of BoundingBox may require a fundamental re-write of BoundingBox to significantly revise or replace the algorithm. In general there is not a direct method for solving for the exact extrema (min/max) of general NURBS objects; iteration is needed. (There are exceptions; for example planar surfaces). One approach would be to use the existing approximate method based on the display/render mesh and determine the corresponding extrema. The refine the mesh determine the corresponding extrema, and compare the result to the previous result. If the difference is not within the desired tolerance the refine the mesh again, etc. If the difference between two levels of mesh refinement is within the tolerance then the extrema of the mesh is likely to be within tolerance of the exact extrema of the NURBS object. To increase the confidence of being within the tolerance a further refined mesh can be used and the result of that mesh tested against previous results.