Abstract : The university Hospitals of Geneva, formed since 1995 with a network of hospitable establishments, with 7'900 collaborators working on 5 geographic sites, with 2185 beds, and offering annually 175'000 hospitalizations (HC: 52'000) and 620'000 ambulatory consultations (2001). The cantonal hospital of Geneva was one of the pioneers of medical data processing in Switzerland, even in Europe. From 1976, a team of IT specialists-doctors, designed the DIOGENE system, which went through numerous developments. At present, the data base of patients DIOGENE gathers data for a little more than a million of patients, with an annual increase of the order of 45'000 new patients. From the beginning, the DIOGENE system was designed as a patient oriented system and not as an administrative system of management. Such a data base could be exploited from the statistical point of view, for example for queries of a medical type or administrative data. During the first years of use, this type of query was made in batch mode, during the ‘slack' periods. However, this type of exploitation gradually met with several obstacles and we envisaged the possibility of creating an “archival” data base with the possibility of directly interrogating an anonymised data base, by using browsers, without using the operational data bases resources. This project lives by the implementation of the data base integrated ARCHIMED system in 1993. The key concept of this application lies in the transcription of the data linked to the patients in a series of "elementary facts", which can later be manipulated in an extremely powerful way. Practically, the pre-selected data from the various operational computer systems, are filtered daily by a series of interfaces, which transforms them into elementary facts then places them in the ARCHIMED data base on an annualised base. After ten years of activity (1993-2003) the annual evolutions are the following : Increase in the neighborhood of 1,5 Gigabytes a year (5MB / j * 365 j. = 1,8 GB) 70 million entries (elementary facts) (i.c. the addition of 200 tables annually) The concept which presided over the birth of ARCHIMED was of a medical and operational order. However, the necessity of harmonizing the data resulting from several different operational systems, forced its designers to an ontological reflection leading to a structuralization meeting the concerns of a conservation of the long-term data. ARCHIMED represent therefore a prototype “historical” data base, designed at the beginning for the management of large amounts of data for long term preservation.