Drought severity and related socio-economic impacts are expected to increase due to climate change. To better adapt to these impacts, more knowledge on changes in future hydrological drought characteristics (e.g. frequency, duration) is needed rather than only knowledge on changes in meteorological or soil moisture drought characteristics. In this study, effects of climate change on droughts in several river basins across the globe were investigated. Downscaled and bias-corrected data from three General Circulation Models (GCMs) for the A2 emission scenario were used as forcing for large-scale models. Results from five large-scale hydrological models (GHMs) run within the EU-WATCH project were used to identify low flows and hydrological drought characteristics in the control period (1971–2000) and the future period (2071–2100). Low flows were defined by the monthly 20th percentile from discharge (Q20). The variable threshold level method was applied to determine hydrological drought characteristics. The climatology of normalized Q20 from model results for the control period was compared with the climatology of normalized Q20 from observed discharge of the Global Runoff Data Centre. An observation-constrained selection of model combinations (GHM and GCM) was made based on this comparison. Prior to the assessment of future change, the selected model combinations were evaluated against observations in the period 2001–2010 for a number of river basins. The majority of the combinations (82%) that performed sufficiently in the control period, also performed sufficiently in the period 2001–2010. With the selected model combinations, future changes in drought for each river basin were identified. In cold climates, model combinations projected a regime shift and increase in low flows between the control period and future period. Arid climates were found to become even drier in the future by all model combinations. Agreement between the combinations on future low flows was low in humid climates. Changes in hydrological drought characteristics relative to the control period did not correspond to changes in low flows in all river basins. In most basins (around 65%), drought duration and deficit were projected to increase by the majority of the selected model combinations, while a decrease in low flows was projected in less basins (around 51%). Even if low discharge (monthly Q20) was not projected to decrease for each month, droughts became more severe, for example in some basins in cold climates. This is partly caused by the use of the threshold of the control period to determine drought events in the future, which led to unintended droughts in terms of expected impacts. It is important to consider both low discharge and hydrological drought characteristics to anticipate on changes in droughts for implementation of correct adaptation measures to safeguard future water resources.