1 7KH$TXDUHOOH5HVRXUFH'LVFRYHU\6\VWHP $ODLQ0LFKDUG a 9DVVLOLV&KULVWRSKLGHV b 0LFKHO6FKROO c,a 0LNH6WDSOHWRQ d  'DOH6XWFOLIIH e $QQH0DULH9HUFRXVWUH a a INRIA - BP 105 - F-78153 Le Chesnay Cedex. b FORTH-ICS, P.O. Box 1385, GR 711 10, Heraklion, Crete, Greece c CNAM, 292 rue St Martin, F-75141 Paris Cedex 03 d System Simulation Ltd., 250M Bedford Chambers, WC2E 8HA, London, UK. e CLRC-RAL, Chilton Didcot, OX1 10QX OXON, UK Aquarelle is a three-year project supported by the Telematics Applications Programme of the European Union, aiming at designing a resource discovery system on the Internet, applied to cultural heritage documentation. The system relies on the Z39.50 protocol to support access to heterogeneous databases, including SGML document repositories. Its most original features are direct linking from SGML documents to database records, an advanced link management facility, and query broadcasting to dynamically selected databases resource discovery; Z39.50; link management; heterogeneous databases; cultural heritage. .  2EMHFWLYHV Many types of resources are made available through the Internet: structured documents (e.g. SGML documents), textual documents with a relatively weak structure (e.g. html pages), unstructured texts (e.g. postscript or pdf files, character files), database records, and various kinds of files containing images, sounds, graphics, etc., which can be considered as unstructured as far as information retrieval is concerned. A (RDS) is a facility offered to end-users to help them to retrieve the resources which are relevant to their current needs. In our view, a RDS should -as far as possible- have the following features: • It should hide the distribution of data from the end user: the Internet (or a sub-part of it) should appear to the user as a global information repository. • It should be able to retrieve structured as well as unstructured information. Information structure, whenever available, should be exploited to allow more accurate retrieval. • It should process information which is organised and indexed to support a retrieval process (e.g. databases), as well as information which is not authored with the objective of facilitating retrieval (e.g. HTML pages on the Web). • It should support heterogeneity of information structure, including heterogeneity of database schemas and of SGML Document Type Definitions (DTD), and this heterogeneity should ideally be hidden to the end-user. • It should support heterogeneity of information sources: HTTP servers, various database management systems, and classical information retrieval systems (such as library OPACs) should be considered as primary content sources. • It should support multilinguality of textual information.