Although recent observations have found evidence of normal galaxies at high redshifts (Cowie et al. 1996; Steidel et al. 1996), there are several reasons to continue to use also high-z radio galaxies (HzRGs) for galaxy evolution studies. For example, since powerful radio sources in the local universe are hosted by elliptical galaxies, HzRGs provide a way to trace the cosmological evolution of ellipticals (the most distant radio galaxy known has 4.4; Rawlings et al. 1996, Nature,383,502). However, the use of HzRGs is heavily complicated because they are not normal galaxies, but active objects where both stellar and non-stellar radiation contaminate each other. This problem becomes critical in the most powerful HzRGs (3C and 4C galaxies), where most of the rest-frame UV light is non-stellar (e.g. Cimatti et al. 1998, ApJ, 499, L21). A possibility to overcome this problem is to use weak radio sources (i.e. with Jy - mJy fluxes at 1 GHz), where the properties of the host galaxies are not contaminated by the AGN component (Dunlop et al. 1996,Nature,381,581). In this regard, weak radio galaxies provide a complementary approach for galaxy evolution studies.
Different types of galaxies contribute to the population of weak radio sources : faint blue galaxies, star forming galaxies, mergers, evolved ellipticals, and quasars. Although it is relatively easy to reach very faint (Jy) radio fluxes, the nature, the redshift distributions, and the relative abundance of the optical counterparts is still an open problem. For example, in the 10-50 Jy regime, discrepant results have been obtained so far: Hammer et al. (1995,MNRAS,276,1085) found a dominant population of high-zellipticals and AGN (at S1.4GHz>16 Jy), whereas Fomalont et al. (1991, AJ, 102, 1258) and Lowenthal (1997, Rev.Mex.A.&A., 6, 105) found a dominant fraction of star forming galaxies. Also, the nature of sub-mJy sources is still unexplored to faint optical magnitudes, and informations have been obtained mainly for relatively bright optical counterparts with B<22 (Benn et al. 1993,263,98) and R<23-24 (Windhorst et al. 1985,ApJ,289,494). The main problem in this research field is the difficulty of reaching very faint optical magnitudes and to obtain large and complete samples in a reasonable amount of observing time. This problem hampers our knowledge of the nature of the faintest sources (often too faint for optical spectroscopy even with a 8-10m telescope), and of the radio-optical luminosity function and its cosmological evolution. In particular, the problem of the cosmological evolution of the luminosity function can be properly investigated only if large and complete samples are available.