With the advent of large galaxy redshift surveys, galaxy classification has gone from many classes (Hubble types) to two! (red and blue sequences). This is partly because we do not have accurate morphologies for 100000 galaxies but also because there appears to be no natural division other than that based on the `galaxy bimodality'. I discuss how recent analyses of the variation in bimodality with stellar mass or luminosity, environment and redshift can provide an empirical picture of galaxy evolution; and present a challenge for galaxy evolution models.