If rainfall, storage and flux divergence are known, evapotranspiration can be estimated as the residual in a budget of atmospheric moisture. We describe a new approach to moisture-budget estimates of Amazon evapotranspiration. Synoptic observations (and numerical model analyses from the U.S. National Meteorological Center) are combined with wind, humidity and rain information retrieved from radiances measured by the Geostationary Operational Environmental Satellite (GOES). Rainfall is inferred from the thermal infrared channel of the Visible and Infrared Spin Scan Radiometer (VISSR). Humidity is retrieved from the VISSR Atmospheric Sounder (VAS). Winds are inferred from both VISSR and VAS images, using visible, thermal infrared and water vapor bands. Satellite and station or model information is merged by means of a recursive filter. The GOES-budget approach was tested over a three-day period from May 1987. Our estimate of evapotranspiration over the basin, 9 mm/d, is unrealistically large. A sensitivity analysis indicates that the largest part of the error originated in GOES cloud and water vapor winds and the next largest part in VAS humidity. We conclude that the approach failed in its first test. However, since the test was conducted researchers have found better algorithms for selecting cloud and water vapor targets and for assigning altitudes. In addition, simulations suggest that even in the absence of better algorithms the next generation of GOES will provide more accurate winds and humidity than the present generation. If it is supported by direct measurements of evapotranspiration, a second test of the approach might be successful. 179 Downloaded by [University of Cambridge] at 06:22 15 June 2016 180 D. W. MARTIN ET AL.