The most visible galaxies (those which have the largest apparent sizes and isophotal luminosities when seen at a given distance) are those with a particular observed surface brightness. Extending this argument to high-redshift galaxies, it is clear that this optimum surface brightness moves progressively to brighter intrinsic surface brightnesses, so as to counteract the effect of K-corrections and cosmological dimming. Thus, the galaxies appearing in faint surveys will be from a population distinctly different from those 'normal' galaxies observed nearby. Galaxies in deep surveys are more likely to be spirals and to be of high surface brightness. This has very important implications for observational studies of galaxy evolution, such as galaxy number counts, redshift surveys and the direct comparison of local and distant samples (e.g., the Butcher-Oemler effect). In particular, these distance-dependent selection biases can mimic evolutionary changes where none exist (or can help to hide real evolution), and can distort the observed luminosity function at high z.