Arsenic contamination of groundwater is posing a serious challenge to drinking water supplies on a global scale. In India and Bangladesh, arsenic has caused the most serious public health issue in the world for nearly two decades. The aim of this work was to study an arsenic removal system based on reverse osmosis at pilot scale treating two different water sources from two different locations in the State of Bihar, India. For this purpose two villages, Bind Toli and Ramnagar in the Patna District were selected, both located very close to the river Ganga. The trials were conducted with aerated and non-aerated groundwater. It is the first time that the arsenic removal efficiency for aerated and non-aerated groundwater by reverse osmosis technology in combination with an energy-saving recovery system have been studied. As the principle of reverse osmosis requires a relatively high pressure, its energy demand is naturally high. By using an energy recovery system, this demand can be lowered, leading to an energy demand per liter permeate of 3-4Wh/L only. Due to high iron levels in the groundwater and as a consequence the precipitation of ferric (hydr)oxides, it was necessary to develop a granular media filter for the trials under aeration in order to protect the membrane from clogging. Two different materials, first locally available sand, and second commercially available anthracite were tested in the granular media filter. For the trials with aerated groundwater, total arsenic removal efficiency at both locations was around 99% and the arsenic concentration in permeate was in compliance with the WHO and National Indian Standard of 10 mu g/L. However, trials under anoxic conditions with non-aerated groundwater could not comply with this standard. Additionally a possible safe discharge of the reverse osmosis concentrate into an abandoned well was studied. It was observed that re-injection of reject water underground may offer a safe disposal option. However, long-term hydrogeological studies need to be conducted for confirmation. (C) 2016 Elsevier B.V. All rights reserved.