This paper shows how to support images in a relational database, so it can fulfill the requirements to be used as the storage mechanism of a PACS. This support includes the ability to answer similarity queries based on the image content, providing fast image retrieval based on indexing structures. The main concept allowing this support is the definition of distance functions based on features, which are extracted from the images as they are stored in the database. An extension to the SQL language enables the construction of an interpreter that intercepts the extended commands and translate them to standard SQL, allowing to take advantage of any relational database server. We describe experiments made with a prototype implemented using these concepts, which allowed answering queries up to 20 times faster than using only existing relational servers.