AbstractThe coupled global climate model (GCM) fidelity in representing upper ocean salinity including near sea surface bulk salinity (SSS) is evaluated in this study, with a focus on the Pacific Ocean. The systematic biases in ocean surface evaporation (E) minus precipitation (P) and SSS are found to be fairly similar in the twentieth century simulations of the Coupled Model Intercomparison Phase 3 (CMIP3) and Phase 5 (CMIP5) relative to the observations. One of the potential causes of the CMIP model biases is the missing representation of the radiative effects of precipitating hydrometeors (i.e., snow) in most CMIP models. To examine the radiative effect of cloud snow on SSS, sensitivity experiments with and without such effect are conducted by the National Center for Atmospheric Research‐coupled Community Earth System Model (CESM). This study investigates the difference in SSS between sensitivity experiments and its relationship with atmospheric circulation, E − P and air‐sea heat fluxes. It is found that the exclusion of the cloud snow radiative effect in CESM produces weaker Pacific trade winds, resulting in enhanced precipitation, reduced evaporation, and a reduction of the upper ocean salinity in the tropical and subtropical Pacific. The latter results in an improved comparison with climatological upper ocean bulk salinity. The introduction of cloud snow also altered the budget terms that maintain the time‐mean salinity in the mixed layer.