This work proposes a comparison of different hydrometeorological reconstruction datasets built on the downscaling of the NOAA 20th century global extended reanalysis (20CR, Compo et al., 2011). It aims at assessing the quality of these reconstructions and improving our knowledge of the multi-decadal hydrometeorological variability over the 20th century. The historical depth of hydrological observations is generally limited to the last 50 years, which is not enough to properly explore the natural hydrometeorological variability, a key to better understand the effects of anthropogenic climate change. In order to overcome this limit, downscaling methods are used to bridge the scale gap between extended large scale reanalyses that are available from the late 19th century onwards, such as 20CR, and local meteorological variables relevant for catchment-scale hydrology. A continuous hydrological modelling using the high-resolution meteorological reconstructions as forcings allows then to reconstruct 20th century streamflow time series. The results from two downscaling methods and two hydrological models are here compared for a reference network of stations in France over the whole 1900-2012 period. The statistical downscaling methods provide either probabilistic or deterministic meteorological fields used as forcings in a conceptual lumped or a physically-based distributed hydrological model. Comparing the different hydrometeorological time series allows (1) quantifying the uncertainties related to the downscaling and the hydrological modelling, and (2) assessing their relative contributions across catchments and across time periods.