The field of climate prediction has been bedeviled by the problem of how to represent the enormous complexity of clouds. The usual strategy is to peform deterministic simulations with advanced cloud models. The study outlined here concentrates on a statistical approach that is arguably better suited to determining the mean climatological state. The presented observations from a wide range of satellite platforms show that a power-law well describes frequencies of occurence of cloud sizes across a very wide range of scales, and that the exponent is robust to local climatological characteristics as surface temperature, aerosol loading, Coriolis forces, or dominant cloud type. Instead, the distribution of cloud sizes emerge simply from a competition for energy and air that occurs due to small-scale cloud mixing processes at cloud edge.