IntroductionThe notion of local (Grothendieck) topos was introduced by Grothendieck and Verdier in SGA 4 [11, VI, 8.4]; a Grothendieck topos % is said to be local if the global sections functor <£-» $f has a right adjoint as well as its usual left adjoint, i.e. if it is the inverse image part of a geometric morphism 5^-> %, as well as the direct image of the unique morphism %^>& > . The example which the authors of [11] particularly had in mind was the topos of sheaves on a space (such as the Zariski spectrum of a local ring) containing a point whose only neighbourhood is the whole space. (Such points have been called focal by Freyd [9]; the similarity between 'focal' and 'local' is deliberate.) They also showed that, for a locally coherent topos %, there is a process of 'localization' which mirrors the passage from the Zariski spectrum of a ring to the spectrum of one of its localizations.Since then, the notion of local topos has not attracted much attention, apart from the occasional passing reference in papers about other things. We believe that the time is now ripe, if not overdue, for a more detailed investigation of the notion, for a number of reasons which we shall now describe briefly.In the first place, the property of being local, like many of the properties of toposes studied in [11], is not so much a property of the topos % itself as of the geometric morphism %-> tf; thus there is scope for studying toposes which are local over some base topos other than the classical topos of sets, or (to change our terminology slightly) studying local maps between toposes. (Note: throughout this paper we shall restrict ourselves to maps of toposes (i.e. geometric morphisms) which are bounded in the sense of [13, Definition 4.43]; we make this restriction in order to be assured of the existence of all the pullbacks which we shall wish to consider, but many of our results are not crucially dependent on it. To emphasize the unimportance of the restriction, we shall denote the 2-category of toposes and bounded geometric morphisms simply by Sop.) Thus, in addition to the properties of local toposes studied in [11], there is scope for studying properties of local maps (stability under composition, pullback, etc.) which cannot even be conveniently formulated in terms of the Grothendieck-Verdier definition.Secondly, the restriction to the locally coherent case in the construction of localizations in [11] seems to have been a matter of necessity rather than desirability (despite the remarks on p. 321 of [11]): Grothendieck and Verdier simply did not have techniques available for handling filtered inverse limits in the non-locally-coherent case. With the more powerful techniques available now, we shall show (in § 3 of this paper) that the notion of localization makes perfectly good sense for arbitrary (bounded) pointed toposes over a base, and we shall also The research of the second author was supported by a Huygens fellowship of the Z.W.O. A.M.S. (1980) subject classification: 18B25. Proc. London Math. Soc. (3) 58 (1989) 28...