Drought in temperate deciduous forests has been shown to decrease transpiration rates in many species; however, total evapotranspiration at the stand level is often invariant to precipitation inputs. We investigate potential causes for this apparent contradiction in a mature oak-hickory forest in North Carolina, USA over four years in which precipitation varied from 934 to 1346 mm yr-1 by combining sap flux measurements from heat dissipation probes with eddy covariance evapotranspiration (ET). We focus on accuracy in sap flux estimates by investigating radial variability in flux rates, nighttime water movement through the stem, and scaling from individual sensors to the entire stand. We found that accounting for nighttime water uptake increased the estimate of growing season transpiration by 26%. ET estimated by scaling sap flux values to the eddy covariance footprint and adding modeled evaporation losses was similar to the eddy covariance-estimated ET (634 ± 26 versus 622 ± 29 mm, respectively). Despite a severe drought and a very wet year, stand evapotranspiration remained consistent across years (s.d. = 26 mm) due to increasing atmospheric demand for water (vapor pressure deficit; D) during dry periods. Thus, whereas it is more difficult for trees to extract water from the soil during drought, days with D are practically absent, enabling most species to maintain relatively high transpiration rates. Furthermore, high D generates increased soil evaporation, compensating for the reduction in transpiration by the most drought sensitive species. Because forest evapotranspiration is relatively invariant to annual precipitation, the consequences of reduced precipitation will be greatest for end users relying on groundwater recharge and streamflow.