Eddy covariance and sapflow data from three Mediterranean ecosystems were analysed via top-down approaches in conjunction with a mechanistic ecosystem gas-exchange model to test current assumptions about drought effects on ecosystem respiration and canopy CO2/H2O exchange. The three sites include two nearly monospecific Quercus ilex L. forests – one on karstic limestone (Puéchabon), the other on fluvial sand with access to ground water (Castelporziano) – and a typical mixed macchia on limestone (Arca di Noè). Estimates of ecosystem respiration were derived from light response curves of net ecosystem CO2 exchange. Subsequently, values of ecosystem gross carbon uptake were computed from eddy covariance CO2 fluxes and estimates of ecosystem respiration as a function of soil temperature and moisture. Bulk canopy conductance was calculated by inversion of the Penman-Monteith equation. In a top-down analysis, it was shown that all three sites exhibit similar behaviour in terms of their overall response to drought. In contrast to common assumptions, at all sites ecosystem respiration revealed a decreasing temperature sensitivity (Q10) in response to drought. Soil temperature and soil water content explained 70–80% of the seasonal variability of ecosystem respiration. During the drought, light-saturated ecosystem gross carbon uptake and day-time averaged canopy conductance declined by up to 90%. These changes were closely related to soil water content. Ecosystem water-use efficiency of gross carbon uptake decreased during the drought, regardless whether evapotranspiration from eddy covariance or transpiration from sapflow had been used for the calculation. We evidence that this clearly contrasts current models of canopy function which predict increasing ecosystem water-use efficiency (WUE) during the drought. Four potential explanations to those results were identified (patchy stomatal closure, changes in physiological capacities of photosynthesis, decreases in mesophyll conductance for CO2, and photoinhibition), which will be tested in a forthcoming paper. It is suggested to incorporate the new findings into current biogeochemical models after further testing as this will improve estimates of climate change effects on (semi)arid ecosystems' carbon balances.