This manuscript gives an overview of several years of research aiming at inventing and evaluating systems and techniques for building efficient content-based retrieval systems managing extremely large databases of multimedia material. It has focused on the management of still images. Most of the work described in this manuscript has been achieved while keeping in mind a database perspective. The most important notion that sheds a particular light on the high-dimensional indexing techniques we have been developing is the \emph{mandatory need for secondary storage}. In essence, secondary storage is needed to cope with failures, which is a requirement for systems running in the real world. Secondary storage is also needed to handle truly large scale datasets.