A PACS database must manage three different types of data; structured data in the form of patient/exam identification information, and unstructured data in the form of text and images. Queries based on the content of text documents as well as the content of images must be suported, in addition to those based on standard, well structured keys such as name, age and sex. We model the PACS database as three logically distinct databases, each supporting one of these data types, with mapping structures relating all three. Several design issues which have a bearing, both on these models and on physical implemtations, are discussed. Because image database systems are the least understood at this time, most of the paper focusses on the them. We briefly discuss current trends in structured and text databases, without discussing commercial systems, and then present some current methods of implementing image database systems.