Horizontal grid-spacings in conventional atmospheric general circulation models (GCMs) are typically between 100 km and 500 km. Hence, many processes are unresolved and must be parameterized in terms of resolved variables. Development of satisfactory parameterizations of mean-field (or domain-average) cloud and radiative processes has been frustratingly slow. Moreover, the sensitivity of conventional GCMs to unresolved interactions between cloud and radiation is unknown. In this study, the native cloud parameterization of a GCM was replaced, in each GCM grid column, by a two-dimensional cloud system-resolving model (CSRM). The CSRMs used a horizontal grid-spacing of 4 km. They were employed to assess the sensitivity of a GCM to the inclusion of interactions between cloud and radiation at unresolved scales as well as sensitivity to accurate domain-average radiative flux profiles. For a simulation spanning one season, unresolved cloud-radiation interactions affect the statistics of average cloud fraction and average cloud-radiative effect, while the accuracy of domain-average radiative flux profiles had less effect.