> I hope you get the idea. That makes only 18 possibilites, which is still 2^18=262 KB and a lot of possible facings. Then there's nowhere to start with handcorrecting the database... Not even to speak about adding support for special things like 1-layer thick wings!

Koen, this isn't a statistical calculation... :)

We already know the spatial model we want to work on: it has 26 faces and we know what kind of normal index belongs to what face.
What is needed, is a way to identify the spatial relationship of a voxel so it kan be linked to one of those 26 faces. For the 26 face model, a 3x3x3 grid will do fine.
If i'm correct, all you need to do is :

-make a loop that walks through all the voxels in the file (for..next loop)

-obtain the coordinates of the first voxel

-check the 26 position in the grid, relative of each concerning voxel.

-If the voxel next to it you set a flag=1, if not then flag=0. Store all 26 flags and have that string then looked up in a table that has the corresponding normal indexes.

-assign that normal index to the voxel

-next voxel...

so, the number of calculations will only increase with the number of voxels in the file...

hell, i have to restrain my self or i start an autonormaliser in Excel :D