The potential benefit of gold nanoparticles (GNP) to radiotherapy has been demonstrated in a range of cell lines and radiation sources as well as in rodent models, sometimes with contradictory results. Few experimental studies have explored the involved deleterious species, hydroxyl radical being so far the most cited, whereas theoretical studies have usually focused on secondary electrons emitted from GNP, making comparison between these two approaches difficult. Here we focus on the physicochemical step (i.e. radical production) and report the first experimental determination of both hydroxyl radicals and solvated electrons yields of formation in the presence of GNP. We also compare these yields between X- and gamma-rays under different atmospheres. Our main finding is a massive and equivalent production of both species under X- and more surprisingly gamma-rays. For concentration as low as 1 nM (0.02% wt of gold), both radiations lead to 3 to 4 times more radicals than water radiolysis without GNP. This is in contradiction with a physical prediction of dose enhancement. Supported by our whole set of experiments the key role of water molecules at the nanoparticle interface in the radical production emerges. This leads us to propose the paramount importance of the physico-chemical step compared to the physical one. Classical approaches based on energy-absorption coefficients and electron ejections should therefore be revisited. (C) 2018 Elsevier Inc. All rights reserved.