Hi Michael, thanks for your reply.

I didn't want to sound arrogant - sorry, my english isn't up to the finer tunes sometimes :-(

I am actually quite aware of how difficult this area is, so I am more looking for a kind of "prefer seam point alignment over poly reduction" or something along this line, not for a "perfect" solution which may be impossible anyway.
For the puzzle object, I created one puzzle-spline and copied it three times, then scaled the middle one smaller, then lofted those three. So the mesh is very clean, since the splines are all three exactly the same, created from one continuous splinedrawing.
But when I fillet the top and bottom with G3 and then export the result, I get this - the points are not aligned:



If I create a default cylinder and fillet the top edge, the polys are aligned perfectly.

So if you should happen to have a clever idea that could reduce this problem I would be very happy. :-)
I do a lot of conversions to polygons for clients lately and this is the main problem I run into.

Another idea to solve this problem: Since a G3 fillet is supposed to have the same surface angle on both sides of the seam, would it be possible to create two small rows of polygons (one on each side of the seam) that are "flat"/have the same normal direction as a buffer zone for the normal-algorithm of the rendering software?
So instead of just angle based calculation, there would be another layer that makes sure that joined edges between NURBS patches get a special treatment?

But maybe it is just XSIs obj import? I'm not sure it handles the imported "user normals" correctly?
Anyone else using XSI?

I'll send you the object via email.

Thank you very much!

Thomas Helzle