The definition of reference conditions (RC) is a crucial step in the implementation of the Water Framework Directive (WFD), as the ecological status is defined as a « slight » deviation from RC, corresponding to no or very low impact. As pristine conditions are unlikely to be encountered in most parts of Europe, the problem is to define a practical level of "very low anthropogenic pressure" that makes the task realistic in most cases. What is a "reference" ? Is it somewhere in the past ? is it a fixed, maximum value ? Rivers are highly dynamic ecosystems, controlled by physical, biogeochemical and ecological processes that ensure a high resilience, and the maintenance of biodiversity. A first conceptual choice will be to focus on the possible causes of alteration of these processes to, evaluate the pressures. Reference sites will be searched in basins where human activities have not significantly altered these processes. The cause of alteration will be searched at the basin, reach and site scale. Possible impacting factors refer to land use, pollutions, artificial structures and channelisation altering hydrology and morphology.RC are then defined as the range of variability of biological parameters observed or expected ? in reference sites, and the reference value of a given metric correspond is the most probable one, i.e. the median or the mean. One important problem will be to integrate the long term variability. Reference conditions have to be defined according to the natural typology, based in France on hydro?ecoregions and Strahler order. In a first step, reference sites have been searched in existing data. A national database implemented at the Cemagref contains the data of the IBGN invertebrates index collected since 1992 on more than 3600 sites. Reference sites were first selected from this database in using two independent approaches : 1) expert selection based on a detailed questionnaire, by the field hydrobiologists; 2) GIS selection, based on land use at the basin and riparian corridor scale, and compared with point?sources pollution. Scientific data from reference sites sampled by the Cemagref were also used. The comparison of these three datasets allowed to define reference values of the IBGN index conform to the WFD requirements for most rivers types. It is now possible to define, as a working hypothesis, a « good ecological status » boundary for the IBGN.However, new reference sites will be searched and sampled in 2005 to, complement the reference network and validate the reference values derived from this first exercise. The methodology, including reference criteria, and example of the results will be presented.