Measurements of cloud radiative forcing in polar regions are less reliable than at lower latitudes because of the difficulty in distinguishing between clouds and ice- or snow-covered surfaces. Scene identification can, however, be improved by using multispectral narrow-band radiances. Comparisons were made between scenes identified by the Earth Radiation Budget Experiment (ERBE) algorithm and those deduced from Advanced Very High Resolution Radiometer (AVHRR) radiances in the Arctic for 4 days in July 1985. It was found that they differ significantly from each other both in geotype and in cloud cover. For instance, regions of clear fractional sea ice according to the AVHRR analysis are assigned as being cloudy over open ocean by the ERBE analysis owing to incorrect specification of the sea ice extent. Zonal averages over 2.5¿ wide bands, of fluxes over clear and cloudy regions and hence also cloud forcing, are determined over the Arctic using radiances measured by the AVHRR on NOAA 9 to identify the nature of the scene and the ERBE radiometer to provide broadband radiances. Results are compared with fluxes and cloud forcing determined solely from ERBE radiometer data. Over the southern portion of the domain, where fractional sea ice was prevalent, the ERBE-based values of net cloud forcing were as much as 50 W m-2 smaller than values determined when the scene identification was based on AVHRR measurements. At higher latitudes the ERBE-estimated cloud forcing was larger than that from the AVHRR, and the magnitude of the difference was smaller. ¿1991 American Geophysical Union