The Earth Radiation Budget Experiment (ERBE) uses bidirectional models to estimate radiative fluxes from observed radiances. The anisotropy of the radiance field derived from these models is compared with that observed with the ERBE scanner on the Earth Radiation Budget Satellite (ERBS). The bidirectional models used by ERBE were derived from NIMBUS 7 Earth radiation budget (ERB) scanner observations. Because of probable differences in the radiometric calibrations of the ERB and ERBE scanners and because of differences in their field of view sizes, we expect to find systematic differences of a few percent between the NIMBUS 7 ERB-derived radiation field anisotropy and the ERBS scanner-observed anisotropy. The differences expected are small compared with the variability of the anisotropy which arises from the variability in cloud cover allowed to occur within the individual scene types. By averaging over groups of 40 ERBE scanner scan lines (equivalent to an average over approximately 2000 km) for a period of a month, we detect significant differences between the modeled and observed anisotropy for particular scene types and Sun-Earth-satellite viewing geometries. For a typical 2.5¿ latitude-longitude region these differences give rise to a bias in the radiative flux that is at least 0.3% for the monthly mean and an rms error that is at least 4% for instantaneous observations. By comparing these fluxes derived using the observed anisotropy with those derived assuming isotropic reflection, we conclude that a reasonable estimate for the maximum error due to the use of incorrect bidirectional models is a bias of approximately 4% for a typical 2.5¿ latitude-longitude, monthly mean and an rms error of 15%. ¿1991 American Geophysical Union