Hydroclimate projections for the 21st century agree on an increase in low flow severity that could undermine current water management practice and require drastic measures for adapting water uses and for sharing resources among different economic sectors (irrigation, hydropower production, etc.). Understanding droughts and low flows is all the more crucial in this current context of adaptation to climate change. A deep knowledge of the major historical droughts in France constitutes a perfect framework to assess the recent observed events but also future events projected by climate change impact studies. Indeed, a fundamental way of adapting to climate change tomorrow is to adapt to climate variability today. This work proposes a reconstruction of precipitation fields in France over the last century built on the NOAA 20th century global atmospheric reanalysis (20CR). This reanalysis provides for the first time an evolution of atmospheric circulation at high temporal resolution since 1871, assimilating surface pressure observations only. The coarse spatial resolution of 20CR however prevents using directly surface variables like precipitation at the scale of a French catchment. The Sandhy (Stepwise ANalogue Downscaling method for Hydrology) statistical downscaling method is used here to bridge the scale gap between 20CR and precipitation relevant for catchment-scale hydrology. This method is based on the idea that similar large scale circulation patterns (predictors) lead to similar local precipitation (predictand). For a given target large scale situation, the dates with the most similar large scale situations in an archive are identified, and the corresponding local scale observations are selected as the targeted local scale predictions. For each target date, Sandhy gives an ensemble of 125 analogue dates and corresponding precipitation values, taking into account equifinality in the optimization of predictor geographical domains. Such an optimization is moreover done independently for 608 climatically homogeneous zones covering France. Sandhy draws here analogue dates from an archive covering the 1958-2008 period for which local precipitation is also available through the Safran near-surface reanalysis. It is then applied to target situations given by 20CR over the whole 1871-2012 period to derive an ensemble of 125 equally plausible gridded precipitation time series. Precipitation data are then monthly aggregated for each time series to compute the Standardized Precipitation Index (SPI) at different time scales and therefore characterize meteorological drought events in a probabilistic way. The SPI critically allows for comparing historical events and their spatio-temporal development. Results allow us to identify severe events for periods with few available precipitation observations, such as the record-breaking 1921 drought. Such an event – much more severe than the recently experienced ones like 1976 or 2003 – may thus serve as a benchmark reference event for adaptation purposes.