Amphibole could potentially be an important host of water in the upper mantle. Moreover, the decomposition of amphibole has been invoked as a possible cause of the lithosphere-asthenosphere boundary. However, amphibole stability has been experimentally studied mostly under water-saturated conditions, which are unrealistic for most of the mantle that may contain only traces of water. Experiments with low nominal water contents yielded controversial results and were properly hampered by problems in controlling water activity. We have solved this problem with a novel experimental approach. We carried out piston cylinder experiments from 900 to 1350°C and 2 to 4.5 GPa using a peridotitic composition coexisting with an excess H2O-N2 fluid phase. The dilution by inert N2 was used to precisely control water fugacity to values realistic for the upper mantle. Numerous reversed experiments were carried out to circumvent problems with the metastable formation of amphibole. Our data show a dual effect of water fugacity on the stability of amphibole. With decreasing water activity, the stability field is simultaneously displaced to lower pressures, and expanded to higher temperatures. This behavior is due to two different decomposition reactions with dehydration involving only solid phases at low temperature, but melting at high temperature. Along a continental geotherm, amphibole will never be stable for typical upper mantle water contents. However, for a mantle containing 150 – 200 ppm of water, traces of amphibole may form in a narrow pressure interval along an oceanic geotherm. Here, amphibole may contribute significantly to bulk water storage, although most of the water still resides in nominally anhydrous minerals. However, even if amphibole is stable in a restricted depth range, it cannot account for the lithosphere-asthenosphere boundary, since decomposition proceeds through a solid-state reaction and does not involve melting for realistic mantle water contents.