Garnet’s stability in arc magmas and its influences on their differentiation were explored experimentally in a typical basalt, andesite, and dacite at conditions of 0.9–1.67 GPa, 800–1300 °C, with 2–9 wt.% added H₂O, and with oxygen fugacity buffered near Re + O₂ = ReO₂ (~ Ni-NiO + 1.7 log₁₀ bars). Garnet did not grow at 0.9 GPa in any of the compositions, even with garnet seeds added to facilitate nucleation. At 1.0–1.2 GPa, garnet grew as thin rims (< 5 µm) on introduced garnet seeds coexisting with dacitic to rhyodacitic liquids at temperatures ≤ 1000 °C. At 1.3 GPa, garnet grew readily with no seeds from 900 to 1100 °C coexisting with liquids ranging from peraluminous basaltic andesite to rhyodacite, and at 1.46 GPa, garnet was stable as hot as 1150 °C in metaluminous basaltic liquid. Garnet grew as a liquidus phase only in the dacite, a composition similar to the average upper continental crust. Inverse experiments on the dacite determined a liquidus multiple-saturation point with garnet, plagioclase, orthopyroxene, calcic clinopyroxene, and amphibole at 975 °C, 1.46 GPa, with 7 wt.% dissolved H₂O. Such dacitic and more evolved melts can be products of peritectic reactions that with decreasing temperature consume garnet, calcic clinopyroxene, and melt components, producing amphibole and less abundant but more evolved melts. For this reason, experiments on product melts need not produce reactant minerals, accounting for some disparities in published experimental results on the apparent stability of garnet in intermediate-to-evolved arc magmas. Results on more mafic compositions are more reliable guides and show that liquids of arc dacitic composition, and more evolved compositions, would coexist stably with garnet only in the deepest portions of continental-margin arc crust with average thickness and density (~ 43 km, ~ 1.2 GPa) or in the underlying shallow mantle. Metaluminous arc basaltic, basaltic andesitic, and many andesitic liquids would not coexist stably with garnet at pressures ranging from the crust to at least the midpoint of the mantle wedge, but results in the literature allow that some andesitic liquids with higher Fe/Mg than common in arcs may also saturate with garnet in the deeper portions of average-thickness continental arc crust.