An international program of intercomparison of radiation codes used in climate models has been initiated because of the central role of radiative processes in many proposed climate change mechanisms. During the past 6 years, results of calculations from such radiation codes have been compared with each other, with results from the most detailed radiation models (line‐by‐line models) and with observations from within the atmosphere. Line‐by‐line model results tend to agree with each other to within 1%; however, the intercomparison shows a spread of 10–20% in the calculations of radiation budget components by the less detailed climate model codes. The spread among the results is even larger (30–40%) for the sensitivities of the codes to changes in radiatively important variables, such as carbon dioxide and water vapor. The analysis of the model calculations shows that the outliers to many of the clear‐sky calculations appear to be related to those models that have not tested the techniques used to perform the integration over altitude. When those outliers are removed, the agreement between narrow band models and the line‐by‐line models is about ±2% for fluxes at the atmospheric boundaries, about ±5% for the flux divergence for the troposphere, and to about ±5% for the change of the net flux at the tropopause as CO2 doubles. However, this good agreement does not extend to the majority of the models currently used in climate models. The lack of highly accurate flux observations from within the atmosphere has made it necessary to rely on line‐by‐line model results for evaluating model accuracy. As the intercomparison project has proceeded, the number of models agreeing more closely with the line‐by‐line results has increased as the understanding of the various parameterizations has improved and as coding errors have been discovered. The most recent results indicate that several climate model techniques are in the marginal range of (relative) accuracy for longwave flux calculations for many climate programs. However, not all such models will give such accuracy. It is recommended that a code not be accepted to provide such accuracy until it has made comparisons to the line‐by‐line results of this study. The data necessary to make such comparisons are included herein. However, uncertainties in the physics of line wings and in the proper treatment of the water vapor continuum make it impossible for the line‐by‐line models to provide an absolute reference for evaluating less‐detailed model calculations. A dedicated field measurement program is recommended for the purpose of obtaining accurate spectral radiance rather than integrated fluxes as a basis for evaluating model performance.