Hi Barrie, so the way Karsten writes about above helps to make higher quality individual surfaces because each surface is more relaxed and isn't trying to go through a significant change of shape in a small area. For a technique like that areas that are tightly curved would be constructed as fillet or blend surfaces, and fillet or blend surface cross section stays with that tight bend as it travels along its length, so even though it has a tight bend in it, its cross section does not rapidly change.

The tricky thing about that technique is that the big broad lightly curved areas need to be built out of larger quadrilateral sheets that then are going to be trimmed to cut off excess area where you want to have a transition in shape. It can be difficult to get this extended shape if you're working off of something like digitized physical data because the extended areas just aren't present in the physical model itself.

If you are constructing something from scratch it can be good to have it set up that way instead though, where each broad section of your shape starts out with an extended, simplified and relaxed surface sort of defining the ideal extended shape of that zone of the surface, and then some type of intersection (either through filleting which has an implied intersection and cutting with it, or by direct manual trimming) cuts away some portion of that extended broad shape to get the final result.

Anyway that's usually the preferred way with NURBS modeling to generate the most relaxed and highest surface quality.

- Michael