I will consider here some practical astrometric requirements, from the perspective of an optical astronomer working on objects much fainter than the limits of the primary catalogues, i.e., from 15 mag down to 20 mag and fainter. I take ‘high density’ to mean surface densities of order 100 or more stars per square degree; these two criteria are roughly equivalent (Bahcall & Soneira, 1980). The positions of such objects are routinely available to an accuracy of around 0.5-1.0 arcsec, e.g. from various digitised versions of the Schmidt Telescope photographic sky surveys. However, these sky surveys themselves readily yield internal accuracies at the 0.1 arcsec level, as demonstrated by repeated measures of the same plates. The problem is that the discrepancies can be ten times larger than this when different machines or plates are used, when a wide magnitude range is covered, when relative positions are compared over distances of a degree or more, or when absolute positions are needed for comparison with, say, radio catalogues. The challenge now is to improve this accuracy; we should to be able to specify the absolute positions of all well-defined optical objects to better than 250 mas, anywhere in the sky, at any epoch and regardless of magnitude or colour.