Well, I think I managed to fix it, at least to a close-enough stage. I want to do some head morphs, so the first step was to make a .obj of just the figure's head. In Poser 3, make every other part of the figure invisible (and don't forget the eyeballs) and export as a .obj -- later versions are a lot easier. The rest was done with a mix of a good text editor and a spreadsheet. The section you need to work on is the vertext data -- all those lines starting "v ". Select and copy to a new file. Make sure that you remove multiple-space sequences. Import into the spreadsheet. Use the single space characters as the markers between columns. You should end up with four columns, one of just "v", and three of numbers. Column C is the y-coordinate, which is where I was seeing the problem. Here I did some guesswork, but you can probably do better by using a morphman.exe program. Use your original .obj file as a morph target, extract it with morphman as a .pmf file, and look at the deltas. What I did, by trial and error and a lucky guess, was to increase the y-coordinates by 0.005 -- this gives delta values in the 5th decimal place. Curiously, they're not consistent, and I suspect that Poser 3 has some sort of binary internal representation which creates rounding errors at this precision. Meanwhile, save back as a .txt file from the spreadsheet, and do a cut and paste to replace the original vertex lines in the .obj file. At least it is only a vertical displacement. I wonder if the notional floor isn't quite at y=0. Hope that's of some help...