This paper presents some problems and solutions concerning the federation of several geographical databases, in the context of interoperability. When a multi-database query is executed, it is important to route it only to sites which are likely to contribute to the answer, whatever the fragmentation might be, geographical, thematic or heterogeneous. Secondly, when the answer implies several spatial databases, it is important that not only boundary alignment is achieved, but also topological continuity. After describing several general problems and concepts about distributed spatial databases and their integration, structures for spatial indexing in multi-database systems are described, giving emphasis to local and global spatial indices. An r -tree-like structure is proposed for organizing those global indices. Geographical fragmentation presents disadvantages to artificially cut geographical objects, such as roads, rivers, etc. In other words, some objects, called fragmented objects, have the particularity of having several database identifiers (one per site) and a mechanism must be provided in order to re-construct those objects when necessary. In this paper, special tools are presented in order to ensure continuity of fragmented objects, semantically (at the level of identifiers), topologically (at the level of data structure) and geometrically (at the level of coordinates, possibly with errors). In order to accelerate queries against fragmented objects, some adjacency tables must be constructed. Truly seamless GIS data will only exist on the Internet once spatial continuity is ensured.