Spectral index distributions can be indicated conveniently by gS v1(αv1,v2), i.e. the distribution of the two-point spectral index between frequencies v1 and v2 for a sample of radio sources complete to flux density S at the selection frequency v1 (flux densities will be expressed in Jy and frequencies in GHz). Such a detailed specification is necessary because g(α) has been found to depend on all three parameters. The effect of varying v1 is well-known: the fraction of flat-spectrum sources increases with selection frequency. Variations of g(α) with S have also been found: both the fraction of flat-spectrum sources and the mean spectral index of the steep-spectrum sources depend on S, at least at high frequencies. The magnitude of this dependence on S as well as g(α) itself appear to depend on the choice of v2, or rather on the relation between v2 and v1. This complex behaviour of g(α) is not unexpected if one considers that the redshift and luminosity distributions of a source sample vary with selection frequency and flux density limit. Intrinsic or induced correlations between spectral index and redshift or luminosity, or different redshift distributions of flat-and steep-spectrum sources may cause variations of g(α) with v and S. An understanding of the behaviour of g(α) requires observations over large ranges of frequency and flux density. Even then, it will be difficult to interpret these data without information about the composition of the various samples, which can be obtained only through optical work. Here we describe recent spectral index information for weak sources selected at 1.4 GHz, as well as work on spectral index/optical identification correlations in 5 GHz samples.