Mixed-phase clouds with both liquid and ice coexisting play a critical role in modulating the Earth’s radiation flux, and precipitation formation. The ratio of liquid water to ice in mixed-phase clouds, largely determined by the presence of ice nuclei and cloud temperature, alters their radiative properties. In this study, the supercooled liquid fraction of mixed-phase clouds is compared among six state-of-the-art global climate models (GCMs) and with NASA Cloud and Aerosol Lidar with Orthogonal Polarization (CALIOP) retrievals. It is found that the GCMs predict vastly different distributions of cloud phase (i.e., liquid versus ice) for a given temperature, and none of them are capable of reproducing the spatial distribution or magnitude of the observed phase partitioning. All the GCMs underestimate liquid water at mixed-phase cloud temperatures, which can affect cloud radaitive forcings, because of the difference in optical properties of liquid versus ice. The sensitivity of the simulated cloud phase in GCMs to the choice of heterogeneous ice nucleation parameterization is investigated. The response to a change in ice nucleation is quite different for each GCM, and the implementation of the same ice nucleation parameterization in all GCMs does not reduce the differences in simulated phase among GCMs. The results suggest that processes subsequent to ice nucleation (e.g., the growth of ice crystals through Wegener-Bergeron-Findeisen process, autoconversion, accretion, and sedimentation) are at least as important in determining phase simulated by GCMs, and thus should be the focus of future studies aimed at reducing uncertainties among GCMs.