Uncertainties in cloud simulations can affect key physical processes in climate modelling. Using satellite data from CloudSat/CALIPSO and CERES, together with ERA5 reanalysis data, this study analyzes the distribution characteristics of global climatological cloud fraction and cloud radiative effects (CREs) simulated by the atmospheric component FAMIL in the climate system model FGOALS-f3 at resolutions of 100, 25, and 12.5 km. Moreover, the study investigates the potential sources of simulation errors based on the Xu–Randall cloud diagnostic scheme. The results show that at different resolutions, cloud fraction is underestimated in low-latitude areas and overestimated in high-latitude areas and that the simulation performance changes nonlinearly with increasing resolution. On the global scale, the deviation in the 25 km model increases compared with the 100 km model, and the cloud fraction deviation of the 12.5 km model is approximately 4% higher than that of the 25 km model in regions such as the Western Pacific Warm Pool. As model resolution increases, the simulated CRE decreases in association with a reduction in cloud fraction. In the Xu–Randall scheme, relative humidity (RH) is the dominant factor controlling climatological cloud fraction. Resolution enhancement further changes RH by affecting variables such as temperature and vertical velocity, thereby indirectly affecting cloud fraction bias. The weak vertical velocity in the 100 km model indirectly leads to underestimation of middle- and high-level cloud fractions in low-latitude areas. The higher temperature in the 25 km model compared with the 100 km model increases the cloud fraction bias. Furthermore, the lower temperature in the 12.5 km model compared with the 25 km model decreases the cloud fraction bias.