This study compares the limb scattered radiances calculated by six radiative transfer models for a variety of viewing conditions. Atmospheres that include molecular scattering, aerosol scattering, and ozone absorption are considered. All models treat single scattering accurately in full spherical geometry. Two approximate spherical models (CDI and LIMBTRAN) rely on the plane-parallel atmosphere approximation to calculate the diffuse radiance field; the remaining four spherical models (Siro, MCC++, GSLS, and CDIPI) treat multiple scattering in a spherical atmosphere. Only three of the models (Siro, MCC++, and GSLS) have vector treatment with polarization. A brief comparison of vector radiances with the limb scattered radiances measured by the SOLSE and LORE instruments demonstrates agreement usually within 15% and always within 30%. The inclusion of polarization appears to have little effect on the level of agreement among the models (which agree to within 2% for this sample case). A more general comparison among calculated scalar radiances follows, including four solar zenith angles (20¿, 60¿, 80¿, and 90¿), three relative azimuth angles (20¿, 90¿, and 160¿), and two surface albedos (0 and 0.95). The single scattered radiances agree to within 1% for almost every case. Comparisons of the total radiance show larger differences, with 2--4% spread among the results of the spherical models. The approximate spherical models show a positive radiance difference relative to the other models that increases with tangent height, reaching as much as 8% at 60 km. The rule used to divide the model atmosphere into discrete layers is shown to affect the calculated radiance, causing a height-dependent difference of up to 1% for 1 km layer thickness.