Yamaoka-san asks about the comparison between the various "large" star catalogues for astrometry. Some months ago I made a simple match between USNO-A2.0 and stars in the Sloan calibration regions, which were done using the USNO-Flagstaff 20cm transit telescope (with CCD). This provides positions near the current epoch for some hundreds of thousands of faint stars near the equator with about 50 milliarcsec (mas) accuracy, i.e. essentially zero error when compared to the scans of photgraphic plates. The result was that the mean errors for fainter stars in USNO-A2.0 were about 0".5 on a star-by-star basis. The systematic errors are quite small.
The GSC v1.2 and Gray's GSC-ACT are very similar. The first is a re-reduction of the original GSC but using the PPM catalogue as the reference net; the second (obviously) is a re-reduction using the better ACT catalogue as the net. Both are very similar in that the sometimes-large systematic errors in the old GSC are removed almost entirely. The _internal_ errors are the same (0".2-0".4) naturally, since no improvement in the scans themselves was made. Given that the mean epoch of the GSC plates is now getting close to 20 years old, current errors in these catalogues has grown somewhat, to perhaps 0".4 on average per-star. Some details about the improvement of the GSC can be found at Jure Skvarc's Web site:
...which if nothing else should convince you that GSC v1.1 should be avoided!
The current catalogue-of-precision for those with reasonably large CCD fields is now Tycho-2. This contains 2.5 million stars and has mean errors for the fainter stars of about 50 mas (0".05). This catalogue supercedes all other such works (SAO, AGK3, PPM, Tycho-1, ACT, TRC), which are now obsolete.
The SDSS calibration regions are described in the "README" file at the USNO-Flagstaff ftp area starting at:
Looking at this just now, I see that this contains 1.2 million stars down to mag. 17 or thereabouts in 16 regions along the equator. If you want to see what your external errors are in reducing CCD astrometry, use these stars, which for most observers will have vanishingly-small errors, i.e. all the scatter you measure will be in your frames, not in the reference net.
In response to a question about doing astrometry of SNe
Star-catalogue choice may be driven largely by star-density rather than by accuracy. As long as you can get at least 9 or 10 suitable stars for reference, then that is enough, especially if those relative few have far higher accuracy than a more populous but lower-accuracy catalogue like A2.0.
If one can get 9-10 GSC v1.2/ACT stars in the CCD frame, these stars should have somewhat smaller errors than USNO-A2.0 for most of the sky, and thus the revised GSCs would be preferable in this case. Note that the GSC contains only about 15 million entries when accounting for the multiple records for the same object, so ~10 stars per 10'x10' frame is typical.
A couple of other points in re SNe astrometry: it might be helpful to obtain average positions from several frames instead of just one, especially if the telescope can be shifted slightly between exposures so that slightly different reference nets are involved, and the location of the target varies somewhat. This will give a better idea of measurement errors than just the fits on the stars in a single image, which can be misleadingly small. Also, I'd like to recommend to those who do this to also measure the nucleus of the host galaxy. Accurate coordinates for the galaxies themselves are often not available, and by providing the Central Bureau with this information, the offsets (and possible problems) can be determined more readily. I have requested that galaxy positions be published on IAUCs with supernova announcements, but without effect. (How many bogus coordinates for SNe have been published?!) Certainly one should not rely on galaxy positions found in most catalogues. New reliable measurements of the galaxy nucleus with current "best" reference stars will always be useful.