Stomatal conductance (gs) impacts both photosynthesis and transpiration, and is therefore fundamental to the global carbon and water cycles, food production, and ecosystem services. Mathematical models provide the primary means of analysing this important leaf gas exchange parameter. A nearly universal assumption in such models is that the vapour pressure inside leaves (ei) remains saturated under all conditions. The validity of this assumption has not been well tested, because so far ei cannot be measured directly. Here, we test this assumption using a novel technique, based on coupled measurements of leaf gas exchange and the stable isotope compositions of CO2 and water vapour passing over the leaf. We applied this technique to mature individuals of two semiarid conifer species. In both species, ei routinely dropped below saturation when leaves were exposed to moderate to high air vapour pressure deficits. Typical values of relative humidity in the intercellular air spaces were as low 0.9 in Juniperus monosperma and 0.8 in Pinus edulis. These departures of ei from saturation caused significant biases in calculations of gs and the intercellular CO2 concentration. Our results refute the longstanding assumption of saturated vapour pressure in plant leaves under all conditions.