As a second step of the international program of Intercomparison of Radiation Codes Used in Climate Models (ICRCCM), an intercomparison of shortwave radiation models was initiated. Among the 26 codes that participated in the comparison were very detailed (line-by-line), narrow-band (high-spectral resolution), as well as highly parameterized (low-spectral resolution) models. A considerable spread was detected in the response of these models to a set of well-defined atmospheric profiles. Substantial discrepancies exist among models even for the simplest case of pure water vapor absorption with standard deviation ranging from 1% to 3% for the downward fluxes at the surface and from 6% to 11% for the total atmospheric absorption. The divergences in downard surface flux increase to nearly 4% when all absorbers and the molecular scattering are considered. In cloudy conditions the divergences range from 4% to 10%, depending on the cloud optical thickness. Another major uncertainty that has been identified is the spectral averaging of the scattering properties which can result in very significant errors for low spectral resolution codes. Since these errors appear to be systematic, they may induce unrealistic feedback mechanisms in numerical climate models. The amplitude of the differences between models is in many cases larger than the accuracy required for the achievements of several objectives of the World Climate Research Program. While reference solutions for the absorption and scattering in atmospheres can be obtained based on the state-of-the-art spectroscopic knowledge and rigorous computational techniques, the absolute tests of the validity of the radiation algorithms would be comprehensive field experiments in which the radiative and all relevant atmospheric parameters are measured to a high degree of accuracy. ¿ American Geophysical Union 1991