In the frame of the QUAE project, an identification procedure was develop to sort singular behaviours in river temperature time series. This procedure was conceived as a tool to indentify particular behaviours in time series despite non continuous measurements and regardless the type of measurement (temperature, streamflow...). Three types of singularities are identified: extreme values (in some cases similar as outliers), roughened data (such as the difference between water temperature and air temperature) and buffered data (such as signals caused by groundwater inflows).