The three-dimensional variability of clouds has an important impact on the transfer of radiation in the Earth's atmosphere, and thus the current practice in climate modeling of assuming clouds are horizontally homogeneous entails significant errors. A new method, based on the backward Monte Carlo technique, has been developed for computing the mean outgoing radiation field from any type of stochastic ensemble of cloud structure. The accuracy of the technique has been verified by comparison with many realizations of deterministic radiative transfer. The method illustrates how spatial variability should be defined, and should lead to improved representations of the radiative effects of clouds in climate models. ¿ American Geophysical Union 1993