This paper describes the architecture and the salient characteristics of the image database system El Nino. The system uses a new interaction model which purports to overcome the problem of the semantic gap. We have a semantic gap when the meaning that the user has in mind for an image is at a higher semantic level than the features on which the database operates. We argue that we can overcome the problems caused by the semantic gap if we replace the usual query paradigm with a more active exploration process, and develop an interface based on these premises. Other relevant aspects of El Nino, which are discussed in the paper are its distributed architecture composed of partially independent engines connected to a mediator, its attempts to integrate visual and textual queries, and its query algebra approach to the problem of putting together queries from different engines. 1 Introduction What is exactly wrong with existing image databases? Most people would wholeheartedly agree that s...