Riparian evapotranspiration (ET) is a major component of the surface and subsurface water balance for many semiarid watersheds. Measurement or model-based estimates of ET are often made on a local scale, but spatially distributed estimates are needed to determine ET over catchments. In this paper, we document the ET that was quantified over 3 years using eddy covariance for three riparian ecosystems along the Upper San Pedro River of southeastern Arizona, USA, and we use a water balance equation to determine annual groundwater use. Riparian evapotranspiration and groundwater use for the watershed were then determined by using a calibrated, empirical model that uses 16-day, 250–1000 m remote-sensing products for the years of 2001–2005. The inputs for the model were derived entirely from the NASA MODIS sensor and consisted of the Enhanced Vegetation Index and land surface temperature. The scaling model was validated using subsets of the entire dataset (omitting different sites or years) and its capable performance for well-watered sites (MAD=0.32 mm day−1, R2=0.93) gave us confidence in using it to determine ET over the watershed. Three years of eddy covariance data for the riparian sites reveal that ET and groundwater use increased as woody plant density increased. Groundwater use was less variable at the woodland site, which had the greatest density of phreatophytes. Annual riparian groundwater use within the watershed was nearly constant over the study period despite an on-going drought. For the San Pedro alone, the amounts determined in this paper are within the range of most recently reported values that were derived using an entirely different approach. However, because of our larger estimates for groundwater use for the main tributary of the San Pedro, the watershed totals were higher. The approach presented here can provide riparian ET and groundwater use amounts that reflect real natural variability in phreatophyte withdrawals and improve the accuracy of a watershed’s water budget.