Independently evaluating large-scale sea ice models is challenging because direct observations are sparse and many satellite products are assimilated into the models being evaluated. Moreover, commonly used sea ice concentration products provide limited information on interior pack-ice properties such as ice thickness. To address these issues, we present a framework for evaluating sea ice models directly in satellite microwave radiometer brightness temperature (TB) space. The framework uses an observation operator to simulate all AMSR2 channels except 7.3 GHz. The observation operator couples emission and radiative-transfer models of the ocean, sea ice, snow, and atmosphere for non-melting Arctic conditions. To reduce simulation uncertainty, the snow-scattering parametrisation and multi-year ice scattering properties were constrained using field and airborne observations. We introduce a multi-channel evaluation metric and demonstrate its sensitivity to errors in sea ice concentration, ice-type fractions, thickness, snow depth, and surface temperature. Application to two versions of a pan-Arctic sea ice model showed that improvements and degradations in modelled sea ice thickness are captured by the metric. The metric was significantly correlated with improvements in first-year ice fraction (<em>R</em><sup>2 </sup>= 0.31) and ice thickness (<em>R</em><sup>2 </sup>= 0.14), but not with snow depth or surface temperature. These results demonstrate that TB-space evaluation provides a complementary and largely independent approach to sea ice model assessment across both the marginal ice zone and interior pack ice.