The Water Framework Directive implementation requires models involving environmental indicators to report water state, to describe anthropogenic pressures, and to assess impacts of restoration measures. These indicators are built from observations, modelling results, expertises and processes hypothesis... They are routinely used in information system by stakeholders and managers. Our work, funded by Rhone-Méditerranée and Corse Water Agency, aims to develop new Pressure/State models to help the choice of restoration measures. In this presentation, we describe the first steps of the approach and focus on the specification of a water quality indicator taking account availability and quality of information. In France, water quality information is providing by different networks. These networks can be permanent or temporary and managed at different levels. Two national monitoring networks are managed by the French water agencies for European reporting and monitoring water quality of French rivers: the first one aims to get a global knowledge of rivers water quality, over the long term; and second one is dedicated to monitor temporarily water bodies with specific issues (degradation, restoration, non-achievement of good status. . . ). These networks are recent (2007) and replaced the former national networks. All the other available water quality data correspond to research or engineering studies. Most of these data only concern one site, cover short periods. Data are gathered in public databases, and are freely available. Water quality databases may include up to 41 physico-chemical parameters plus 4 biological parameters. These databases are heterogeneous because data are issued from different networks, using evolving measurement locations and changing sampling and analytical protocols. Moreover, these observations depend on environmental conditions, and anthropogenic pressures. By definition, an indicator has to be enough robust to be compared, spatially and temporally if ones wanted to summarized it by mean value. We question the meaning of mean values to summarize the water quality and study in which conditions it is relevant to use such values. We focus on a biological indicator (IBGN) because of its robustness and its wide use. It is based on the abundance and the diversity of the benthic invertebrates. The work is applied to the Saône catchment (30000 km2), located on the East of France. We perform an extensive gathering of available data to evaluate the relevance of the use of mean value on our series. On the Saône catchment, there are 191 stations belonging to the two national monitoring networks, 14 stations belonging to the former national monitoring networks and 1619 stations issued from research or engineering studies. Altogether 2534 IBGN observations are available from May 1993 to October 2008. Yet there are only 56 stations with more than 10 observations, i.e. a total of 693 IBGN observations. Between 10 and 22 measures are available for each station that all belong to the national monitoring networks. In order to test the signification of mean values, we study series' stationarity. We used four different tests (Spearman, Mann-Kendall, Permutation and linear regression) to detect trends and breaking points. The results are not exactly the same depending on the test: some series are qualified as stationary by one test and not stationary by another one. This result allowed comparing test efficiency for short temporal series. Nevertheless, we demonstrate that the upstream part of the Saône River, the Doubs River as well as other small tributaries of the head of the catchment show a significant water quality trend. Hence the mean value is a bad indicator of the water quality for more than half of the stations. We conclude by a discussion about trends explanation, and a proposal of new indicators.