Phase relationships in natural andesitic and synthetic basaltic systems were experimentally investigated from 2.2 to 7.7 GPa, and 550°C to 950°C, in the presence of an aqueous fluid, in order to determine the stability of hydrous phases in natural subducted crustal material and to constrain reactions resulting in the release of water from subduction zones to the mantle wedge. Water reservoirs in subducted oceanic crust at depths exceeding the amphibole stability field (>70–80 km) are lawsonite (11 wt % H2O), Mg‐chloritoid (8 wt %), talc (5 wt %), and zoisite‐clinozoisite (2 wt %) in basaltic rocks; and lawsonite, zoisite‐clinozoisite, phengite (4 wt %) and staurolite (2 wt %) in andesitic compositions. The thermal stability of lawsonite at 6.0 GPa extends to ≈800°C and 870°C in basaltic and andesitic compositions, respectively. At pressures above amphibole‐out (2.3–2.5 GPa) lawsonite reacts through continuous reactions with steep positive dP/dT slopes to zoisite‐clinozoisite (until 3.0–3.2 GPa), and at higher pressures (to more than 7.7 GPa) to assemblages containing garnet + clinopyroxene and garnet + clinopyroxene + kyanite in basaltic and andesitic compositions, respectively. On the contrary, the breakdown of zoisite‐clinozoisite is mainly pressure‐sensitive. Phengite represents the hydrous phase with the largest stability field encountered in this study. In andesite, phengite is stable to more than 7.7 GPa and more than 920°C. Talc and staurolite contribute in minor amounts to the water balance in basaltic and andesitic rock compositions. A model for water release from the subducted slab is developed combining thermal models for subduction zones with the experimentally determined phase relationships. Up to 1 wt % and 2 wt % H2O in basaltic and andesitic rocks, respectively, can be stored to depths beyond 200 km in cold subduction zones, mainly by lawsonite and phengite. Dehydration rates are high until amphibole‐out, and relatively low at greater depths. The amphibole‐out reactions are found to release a significant amount of water in a depth interval of several kilometers, however, they do not represent a discrete pulse of fluid and do not completely dehydrate the descending slab. Fluid release at depths greater than 200 km through phengite and progressive lawsonite breakdown would hydrate the overlying mantle, causing the generation of amphibole or phlogopite peridotite. At higher geothermal gradients, epidote/zoisite contributes to fluid flux to the mantle wedge at 100–120 km depth. The extensive stability field of phengite may greatly enhance the role of sediments and the small amount of potassium in mafic compositions for the fluid budget in subduction zones at increasing depth.