Extreme properties are often sought for in astronomy as one way to sharpen our understanding of our most fundamental concepts. In such a context, metallicity proves to be of crucial importance. Because metals build up as a function of time after the release of nucleosynthesis products, the most metal-poor galaxies help us to understand the primordial Universe and the subsequent formation and evolution of galaxies.
We have discussed in this paper the reason why local metal-poor galaxies are found amongst the dwarfs. Whether this property remains true at high redshift where proto-galaxies begin to form, is certainly a challenging question. If the hierarchical model of galaxy formation is correct and if building block galaxies are similar to the well-studied dwarfs we discussed in this review they may be unreachable observationally, even with the advent of the most recent large telescopes. On the other hand, large, massive and gas-rich proto-galaxies may have some properties similar to those of the most unevolved objects we see today. The same applies to the distant starbursts that have been found using Lyman break techniques or Ly emission searches.
Evolution versus formation has been a clear issue in order to characterise the few known extreme star-forming dwarfs like IZw18. Most of the properties of these galaxies: blue colours, gas and dust content, and extreme metal deficiency led to the belief that, although rare, small condensations were able to produce genuinely young galaxies at the current cosmic epoch. This question has indeed received a lot of attention in the recent years but we consider it as somewhat ``passé''. In most cases - but for a few intriguing ones - an old underlying stellar population has been found, revealing one or several previous bursts of star formation. The same seems to apply to dwarf gas-poor galaxies as one can tell from the study of their colour magnitude diagrams. The problem of the existence of local young galaxies may not be completely solved yet but its importance is less acute, since in any case, there is no significant population of dwarf galaxies still in the process of formation. Evidences that most of the galaxies were formed at ealier epochs come from a more direct view of what the Universe looks like at high redshift and also the hint that the metal production rate in the Universe has passed its peak. The question of present cosmological interest would rather be to determine the epoch when the bulk of dwarfs were formed. Was the formation of dwarfs confined to very high redshifts, or was it still going on at intermediate redshifts where the faint blue galaxies emerge and the Universe was about 2/3 its present age?
Nevertheless, for years, local extremely metal poor galaxies turned out to be our best test laboratory. We have learned a lot about the properties of their massive stars (formation and evolution, appearance of WR stars), the evolution of the dynamics of the gas in the gravitational potential of the parent galaxy as a superbubble evolves and the chemical enrichment of the interstellar medium after the fresh products are well mixed. We have discussed their dynamical properties and interactions with companions in order to understand the triggering mechanism that ignites their bursts of star formation. These galaxies have been modelled to constrain the scenario of the starbursts, with the major motivation of describing how various classes of dwarfs could be linked together into one or several evolutionary scenarios. This is not completely settled yet and a lot of work is still needed both theoretically and observationally in this area. Surveys are aimed to build luminosity functions that will help in answering some of the above questions and also produce new targets that will open the possibility to investigate the properties of metal-poor galaxies at much larger redshift. We clearly enter a new area in which metal-poor galaxies and the sub class of metal-poor dwarf galaxies will observationally begin to be considered in cosmology. They enter in the study of the cosmic metal enrichment, the possibility to establish new distance indicators, the study of the yield of heavy elements and the derivation of the primordial helium, and finally the rôle these galaxies play in the hierarchial buildup of structures in the Universe.
Will we ever find a metal-free galaxy? In our local universe we concur that young galaxies (hence possible metal free candidates) are unlikely to be found. In fact we have stressed that the most metal poor known dwarf galaxies are orders of magnitudes more metal rich than the most extreme halo stars in our Galaxy (Cayrel 1996). The metallicity distribution of the Galactic disc stars and the so-called G-dwarf problem (Pagel 1997) indicate pre-enrichment before the disc formed. The oldest known systems in our Galaxy (and the Universe), the halo globular clusters, have metallicities similar to local dwarf spheroidal galaxies. Hence, the oldest globular clusters were self enriched or formed from pre-polluted material. At high redshift the challenge is to pick up galaxies at the epoch of formation. If our Galaxy is a representative case we do not expect pristine (or nearly pristine, with abundances much lower than the most metal-poor local dwarf galaxies) matter to be found anywhere but in dwarfs or sub galactic clumps.