Context. Detection of molecular oxygen and prediction of its abundance have long been a challenge for astronomers. The low abundances observed in few interstellar sources are well above the predictions of current astrochemical models. During the Rosetta mission, an unexpectedly high abundance of O 2 was discovered in the comet 67P/Churyumov-Gerasimenko's coma. A strong correlation between O 2 and H 2 O productions is observed, whereas no such correlation is observed between O 2 and either of CO or N 2 . Aims. We suggest that the O 2 molecule may be formed during the evaporation of water ice. We propose a possible reaction:, a molecule which should be co-produced during the water ice mantle growth on dust grains. We aim to test this hypothesis under realistic experimental conditions. Methods. We performed two sets of experiments. They consist of producing a mixture of D 2 O and D 2 O 2 via the reaction of O 2 and D on a surface held at 10 K. The first set is made on a silicate substrate, and explores the limit of thin films, in order to prevent any complication due to trapping during the desorption. The second set is performed on a pre-deposited H 2 O ice substrate and mimics the desorption of mixed ice. Results. In thin films, O 2 is produced by the dismutation of H 2 O 2 , even at temperatures as low as 155 K. Mixed with water, H 2 O 2 desorbs after the water ice sublimation and even more desorption of O 2 is observed. Conclusions. H 2 O 2 , synthesised during the growth of interstellar ices (or by later processing), desorbs at the latest stage of the water sublimation and undergoes the dismutation reaction. Therefore an O 2 release in the gas phase should occur at the end of the evaporation of ice mantles. Temperature gradients along the geometry of clouds, or interior of comets, should blend the different stages of the sublimation. Averaged along the whole process, a mean value of the O 2 /H 2 O ratio of a few percent in the gas phase seems plausible.