[This is a very belated follow-up to my original post where I described how to test autofocus. More specifically, these are the results from testing my 20D.]
The 20D had trouble latching on to focus even with the florescent desk lamp aimed at the paper as shown in the setup image at left. Overall, I'd say the 20D doesn't seem to acquire focus as fast and as in as low light as my 350D does, even though the 20D is supposed to have a better focusing system (one stop more sensitive, -0.5 EV vs +0.5 EV). That's just my general observation though, and I've never really tested it scientifically.
Either way, for the this auto focus test, I found it very helpful to aim a flashlight at the center of the focusing target. With the flashlight it latched on right away and didn't hunt at all.
To get the autofocus to search both ways, I held my hand in front of the lens (so it had to hunt) until it was on the extreme of the direction that I desired to test, and then let if find the focus. I also did a manual focus test, but I only really got good manual focus results if I took my time and moved the focus back and forth repeatedly. Just trying to find the focus quickly gave pretty horrible results.
The camera was mounted on a tripod of course.
For the measurements, I tested my Tamron 17-50mm f2.8 at 50mm and my Canon 70-200mm f/4L USM at 200mm. Obviously, a more complete test would be nice, but I didn't want to take a lot of time to determine if there was a problem with my camera. For each lens I took three measurements for each of manual focus, autofocus from near (close focus) to the desired and from far (infinity focus) to the desired. In each case I also computed the degree of field using this DOF calculator. All measurements are in millimeters.
Canon 70-200mm F/4: 200mm, 1/25s, F/4, ISO 100
+- 7mm DOF
Manual: +2, +18, -12
From Near: +4, -2, -6
From Far: +1, -1, -3
Tamron 17-50mm F/2.8: 50mm, 1/40s, F/2.8, ISO 100
+- 7mm DOF
Manual: -17, -10, +5
From Near: -9, -11, -14
From Far: -13, -12, -13
For the Canon 70-200mm F/4, autofocus seems spot-on, and in fact better than I can do manually most of the time. So no problems there.
For the Tamron, autofocus gives about 1.5x degree of field, which is on the border of a miscalibration (especially since it consistently focuses too close by about 10 mm). With an F/2.8 lens, it will use the special autofocus sensors in the 20D, so it may be a slight misalignment with those. Still though, without a warranty and an obvious problem (I'd have a hell of a time explaining to SquareTrade that my measurements indicated a problem) it isn't worth pursuing. As long as I use a little more degree of field than I need in my shots, I should be fine.
My viewfinder plane calibration seems to be fine, because my manual focused images are consistent (no definite close or near focus problems).
I also tested each sensor to find the offset in my camera and verify operation, and everything was consistent with the published information. The highly scientific technique involved me drawing lines with a pen, freehand, on my target and selecting different autofocus points to see if they could acquire focus (the image that opened this post shows the pen lines). I drew both thin (one line-width) and thick (two or more passes with the pen) lines because I had some trouble finding the thin lines.
On the Tamron (which uses the enhanced sensor), it found the thin vertical line every time, but missed the horizontal thin line every time. It got the horizontal thick line though. All other sensors (the F/5.6 ones) got the thin lines every time and worked as expected. On the Canon, it was able to get all the thin lines, but of course the magnification was larger.
All sensors seemed to have the correct orientation, but the full sensor field seemed to be a bit higher than shown on the viewfinder screen (a very common alignment problem). Nothing to worry about though.
So, that's it! The whole process (and research that went into it) kind of burst my bubble about how accurate (rather, inaccurate) auto focus is, but I'm happy to know that my 20D doesn't have any major problems. I've heard so many horror stories of cameras going in for service and coming back with more problems (and the autofocus isn't fixed!).