Solar radiation assessment by satellite is constrained by physical limitations of imagery and by the accuracy of instantaneous local atmospheric parameters, suggesting that one should use simplified but physically consistent models for operational work. Such a model is presented for use with GOES 8 imagery applied to atmospheres with low aerosol optical depth. Fundamental satellite‐derived parameters are reflectance and cloud cover. A classification method applied to a set of images shows that reflectance, usually defined as upper‐threshold Rmax in algorithms assessing cloud cover, would amount ∼0.465, corresponding to the transition between a cumuliform and a stratiform cloud field. Ozone absorption is limited to the stratosphere. The model considers two spectral broadband intervals for tropospheric radiative transfer: ultraviolet and visible intervals are essentially nonabsorbing and can be processed as a single interval, while near‐infrared intervals have negligible atmospheric scattering and very low cloud transmittance. Typical values of CO2 and O3 content and of precipitable water are considered. A comparison of daily values of modeled mean irradiance with data of three sites (in rural, urban industrial, and urban coastal environments), September–October 2002, exhibits a bias of +5 W m−2 and a standard deviation of ∼15 W m−2 (0.4 and 1.3 MJ m−2 for daily irradiation). A comparison with monthly means from about 80 automatic weather stations (covering a large area throughout the Brazilian territory) still shows a bias generally within ±10 W m−2 and a low standard deviation (<20 W m−2), but the bias has a trend in September–December 2002, suggesting an annual cycle of local Rmax values. Systematic (mean) errors in partial cloud cover and in nearly clear‐sky situations may be enhanced using regional values for atmospheric and surface parameters, such as precipitable water, Rmax, and ground reflectance. The larger errors are observed in situations of high aerosol load (especially in regions with industrial activity or forest or agricultural fires). The last case is evident when sites in the Amazonian region or São Paulo city are selected. When considering daily values averaged within 2.5° × 2.5° cells, the standard error is lower than 20 W m−2; present results suggest an annual cycle of mean bias ranging from +10 to −10 W m−2, with an amplitude of ∼10 W m−2. These values are close to the proposed requirements of 10 W m−2 for the mean deviation and 25 W m−2 for the standard deviation. It is expected that the introduction of a reference grid containing mean values of parameters within a cell could induce a decrease in the standard deviation of mean errors and the correction of their annual cycle. A model adaptation for assessing the effect of high aerosol loads is needed in order to extend improvements to the whole Brazilian area.