We conduct a global assessment of the spatial heterogeneity of cloud phase within the temperature range where liquid and ice can coexist. Single-shot CALIOP lidar retrievals are used to examine cloud phase at scales as fine as 333 m, and horizontal heterogeneity is quantified according to the frequency of switches between liquid and ice along the satellite’s path. In the global mean, heterogeneity is greatest between $-$15 and $-$4 \textdegree C with a peak at $-$5 \textdegree C, when small patches of ice are prevalent within liquid-dominated clouds. Heterogeneity “hot spots” are typically found over the extratropical continents, whereas phase is relatively homogeneous over the Southern Ocean and the eastern subtropical ocean basins, where supercooled liquid clouds dominate. Even at a fixed temperature, heterogeneity undergoes a pronounced annual cycle that, in most places, consists of a minimum during autumn or winter and a maximum during spring or summer. Based on this spatial and temporal variability, it is hypothesized that heterogeneity is affected by the availability of ice nucleating particles. These results can be used to improve the representation of subgrid-scale heterogeneity in general circulation models, which has the potential to reduce longstanding model biases in cloud phase partitioning and radiative fluxes.