Well, of course, we all use them -- and love them. Boy, would we be less productive without them! But sometimes, it's necessary to apply a little skepticism and common sense when one of the databases presents what looks like a big new discovery.
For example, in January, 2000, there was a flurry of excitement over a pair of carbon stars which seemed to show F-star spectra (which would be really wierd). Brian Skiff, who knows a lot about the history of astronomy, did some investigation. He sent these two messages to VSNET-chat (I've edited them slightly -- MWR):
Before getting too excited about the F-type spectra reported for the two southern carbon stars, and drawing from them unwarranted astrophysical interpretations, be sure to look for the most likely cause of the anomaly: human error. In the case of FK Pup, the reason it has an 'HD' type without an HD number is that it was classified by either Cannon & Mayall at Harvard or by folks at McCormick Observatory (Vyssotsky et al.) as part of the Yale zone catalogue astrometric work many years after the original HD catalogue and its extensions. This stuff was not published except in the Yale zone volumes themselves. Having gone over many regions containing these stars comparing them with more recent objective-prism surveys, I can assure you the types are none too reliable. The stars they examined were often underexposed and/or out-of-focus on the plates, and thus the classifier would frequently give a type based on little more than the general shape of the spectrum as a function of wavelength (rather than working from line ratios as would be the correct method). Thus gross errors are common in the set of classifications done for the Yale zones.
Another significant factor relating to the carbon stars is that since the classifications were done on blue-sensitive plates, most likely the carbon stars were too faint and simply not present at all on the plates used. Yet the classifiers knew there was a zone-catalogue star close by, so they tried to find something to fill the bill. I'd be willing to bet in both cases under discussion that within a few arcminutes there is a decently bright star of roughly F type---in other words, the classificiation is correct but it is for some other star....
I just looked in SIMBAD around FK Pup, and sure enough not too far away is GSC 7133-3997, which is reported in the Tycho catalogue a full magnitude brighter (in blue!) than the carbon star and with B-V color 0.4 --- hey, that's just the color of an F star! Now look at the field of CGCS 2792: just 20" away is a blue star (GSC 8960-1845 = PPM 358078) that's 0.7 mag. brighter (in blue) than the carbon star. Indeed SIMBAD duly assigns the 'F0' type to this star, not the carbon star, so somebody's noticed this one before. QED. (I don't insist on collecting on my bet.)
In re the IRAS LRS classifications: as Mr. Greaves notes, the astrophysical zoo is more complicated than what was possible to discern reliably from the IRAS data; nearly all the LRS spectra are quite low S/N. Thus the machine-derived 'region' types are little more than a guide, and frankly are best ignored. All the objects he mentions have substantial bibliographies shown in SIMBAD, including post-IRAS follow-up spectroscopy, and there is no ambiguity about what they are.
In re visible-light identifications in star catalogues: the GSC can be useful for such searches, but more often you'll want to use the USNO-A series catalogues, since this contains half a billion stars, goes something like seven magnitudes fainter, and includes some star-color information. USNO-A2.0 can be searched using the Strasbourg VizieR utility (catalogue i/252 in their system). There are plenty of stars of extreme color not in A2.0, so be careful even here.
At the risk of possibly boring folks on the list---there's at least three of us interested in it!---here are some more comments about the pre-War spectral surveys following on John Greaves' and Mati Morel's responses. I'll save some remarks about "fixing the databases" for another post.
Although I based my comments about the Yale southern zone work on my own comparison with modern surveys, one can also consult Nancy Houk's remarks on the subject. Houk is about halfway through reclassifying all the HD stars on the MK system (the data for stars south of -12 Dec are published in four volumes). There's a presentation by her starting on page 70 of IAU Symposium no. 50, "Spectral Classificiation and Multicolor Photometry", Fehrenbach and Westerlund, eds., 1973. The paper is a first discussion of progress on the reclassification project. In it Houk makes various comparisons with the original HD data. As part of that comparison, she also classified stars on five of the plates Cannon used for the HD itself. These plates would have had Cannon's marks on them as to exactly the images examined. Houk notes: "It is difficult to overstate the problems with focus on the HD plates. Only the central region of each plate is in good focus, and Miss Cannon classified many stars which were quite out of focus. On some no lines at all could be seen. Consequently the HD type may be uncertain even for strongly exposed spectra." Given that the non-HD Yale stars are roughly 1 or 2 magnitudes fainter than the HD stars, underexposure would have been an additional problem.
All spectral surveys before World War II and many afterwards were done using astrographs with an objective-prism attached. The fields of these lenses are usually not flat, so only part of the field would be in good focus. In addition, because of the chromatic aberration of the lenses, only a narrow range of wavelengths in the spectra would be in focus. The usual workaround would be to take a set of plates at the minimum focus, typically near H-gamma (4340A), which would show most of the important lines for classification. Another set would be taken with the focus adjusted so that the CaII H and K lines toward the ultraviolet (between 3900 and 4000A) would be in focus. This set would allow the H&K strengths to be estimated (or measured), and allow the classifier to examine the Balmer decrement (a luminosity criterion for hot stars). As a result of the color curve of the lens, this second set also often had a region near H-beta in focus, including the TiO bandhead near 4700A, which would be useful for red stars.
There were other problems with the old astrographic lenses. Most of them had generally slow f/ratios. The very largest ones, such as the excellent astrographs at Crimea and Stockholm, were a mere 40cm aperture, huge beasts even so! And apart from the general chromatic aberration problems, in some cases the combined thickness of the glass in the objective (triplet) and the wedge of glass in the prism would mean most of the light blueward of 4000A was soaked up, and only on bright stars were the important H&K lines recorded. This led to B-type stars being misclassed as type F, since their hydrogen lines are of similar strength (the early-B stars will show no H&K, except for sometimes a weak interstellar K line, whereas in F stars H&K are quite strong), and a moderately-reddened B star will have about the same overall spectral curve as an unreddened F stars (and remember before 1930 there was no convincing evidence of reddening at all!). Plenty of B stars are given as F (none of the reverse, luckily, that I've found) in the HD and HDE, in the Bergedorf Durchmusterungen, in the Stockholm and Crimea surveys (which are mostly reliable otherwise), and even in some of the surveys done at Case (which are generally the best quality---taken with a fast Schmidt camera).
I'm sure the folks who did the fainter non-HD stars appearing in the Yale zone lists did the best they could in the circumstances. Just as now, they doubtless had a limited budget and time, and not all the plates they had to work from were the best. Probably also they considered that the types would be used not on a per-star basis but only in a statistical ensemble, so a small percentage of misclassified stars wouldn't ruin such analyses.
My opinion is that the old spectral surveys retain their value, and have even gained a bit of late. Wherever these surveys cover, all the Tycho stars and then some are included, often going two magnitudes fainter than Tycho, nearly to the bottom of the GSC for some areas. The main problem has been that the stars are mostly identified only on charts, so their positions are not readily recoverable without some software and some rather detailed work well beyond the tedium-tolerance of most people (it isn't "science" after all, only bookkeeping). As some of you know, I have gone through a number of the spectral surveys to obtain accurate positions and identifications (which are indispensable for linkage to external lists like IRAS, x-ray/UV surveys, etc.). If you're looking for long lists of spectral types that are not found at the data centers, I've got around 25-30,000 of 'em stashed at the Lowell ftp area ( ftp://ftp.lowell.edu/pub/bas/starcats). Much of this has made its way into SIMBAD thanks to the efforts of the folks in Strasbourg, particularly Fabienne Woelfel. Almost as much as adding reliable stuff to the database, the identification work also reveals various errors in SIMBAD itself. The main value in going through these lists star-by-star may well be to turn up (and fix!) such problems.
This leads to Mati's comments about how to fix the databases, which I'll discuss in another post.