Evapotranspiration by phreatophytes (primarily saltcedar) was determined by the water-budget method for 5,500 acres (2,230 ha) of the Gila River flood plain in southeastern Arizona. The water budget consists of 12 components including surface and subsurface flow through the study area, precipitation on the area, and soil-moisture changes in the unsaturated soil profile. Nine years (1963-71) of hydrologic data were collected on four reaches within the area. These data provided over 400 measurements of evapotranspiration for two-or three-week periods. Midway through the study the vegetation was removed from the flood plain. The evapotranspiration measurements are therefore defined for both natural vegetative cover and essentially bare-ground conditions. This report shows how each component of the water budget was evaluated, demonstrates the significance of each component in relation to the total evapotranspiration, and describes the methods used to evaluate the relative accuracy of each component. The two most significant components of the water budget are, generally, the Gila River inflow and outflow. One of the least significant is tributary inflow, which occurred only 4 percent of the time during the 9-year study. Soil-moisture change is highly significant during periods of low streamflow and is one of the more difficult components to measure. The groundwater flow components are the least variable in the water budget, fluctuating only in response to seasonal changes in the downvalley groundwater slope. The total measurement error of each component consists primarily of a sampling error which is dependent on the number of observation points used to measure the component. This error is time variant, reflecting both the variability in repetitive measurements and the error due to missing data. Included in the total measurement error is a bias error which gives a constant overestimate or underestimate of the component. Only the groundwater flow components introduce a measurable bias error, but the direction of this error is unknown and its magnitude in relation to evapotranspiration is relatively insignificant. The total measurement error in evapotranspiration is not related to the magnitude of evapotranspiration but rather to the total volume of water moving through the area. Thus, the minimum errors occur during the midsummer months of maximum evapotranspiration when streamflow is low and precipitation is negligible. Evapotranspiration rates computed for reach 1 indicate that phreatophyte clearing reduced summer rates by nearly 45 percent. The average computed measurement errors in summer evapotranspiration rates, before and after clearing, are ±59 percent and ±113 percent, respectively, and the average measurement error in the change in summer evapotranspiration as a result of clearing is nea^y ±200 percent. These large computed measurement errors are shown to overestimate substantially the true measurement variable in evapotranspiration. The computed errors do give, however, a good indication of the relative sign...