We attempt to explain the experimental and molecular dynamics simulation evidence that suggests that the freezing of atmospheric aerosols occurs beginning at the droplet surface. By using the capillarity approximation, we derive the reversible work of formation of a crystal nucleus in the cases where it forms homogeneously within a (supercooled) bulk liquid and where it forms "pseudoheterogeneously" at the surface. Comparing the works of formation in these two cases, one obtains a condition that must hold in order for pseudoheterogeneous (surface) crystallization to be thermodynamically more favorable than homogeneous (bulk) crystallization. This condition is satisfied when at least one crystal facet is only partially wettable by its own melt.