The Gokosha autocollimators we used project a fairly narrow beam, so only uses glass relatively near the axis of the lens. If a wider-aperture test lens had, for example, some spherical aberration, it seems to me that the best effective focus point won't necessarily agree with the autocollimator. So although an autocollimator (the ones I'm familiar with) is much more precise than anything else I have, I think the final arbitrator of image focus has to be an actual image on film or sensor.
Originally Posted by Leigh B
I can come up with possible explanations of how a split-image screen might have focus discrepancies compared with either an autocollimator or an actual image, I'm just not sure how to explain the significant variation in the test I mentioned, where everything but the viewer stayed the same.
A ground glass seems to me to be the best way to mimic film (or sensor) response because it accepts image-forming light from all zones of the lens. Assuming that the ground glass position matches that of the film plane. This would certainly be a good test application for an autocollimator with focusing lens.