Long historical series of streamflow are a precious source of information in the context of hydrological studies, such as research of trends or breaks due to climate variability or anthropogenic influences. For this kind of studies, it could be very important to go back as far as possible in the past, in order to highlight information content of historical observations. During our research we concentrate on the Durance watershed (14000 km2) in order to understand last century (1900-2010) hydrological variability due to climate changes and/or anthropogenic influences. This watershed, situated in the Alps, is characterized by variable hydrological processes (from snowy to Mediterranean regimes) and a wide range of anthropogenic influences (hydropower generation, irrigation, industries, drinking water, etc.). We are convinced that this research is necessary before any climate and hydrological projection. Documentary researches lead in collaboration with a historian allowed to find about ten long streamflow series from the beginnings of the 20th century on the Durance watershed. The analysis of theses series is necessary to better understand the natural hydrological behavior of the watershed, before the development of most of the anthropogenic influences. If the usefulness of such long streamflow series is obvious, they have some limitations, one of them being their heterogeneity, which can have many origins: shift of the gauging station, changes in the anthropogenic influences, or evolution in the methods used to build the series. Before their interpretation in terms of climate or land use changes, uncertainty estimation of historical streamflow records is therefore very important to assess data quality and homogeneity over time. This paper focuses on the estimation of the historical streamflow records uncertainty due to the evolution of their construction methods.