As resolutions of nonhydrostatic mesoscale models become increasingly fine, small-scale convective cloud fields are becoming increasingly resolved explicitly. Unlike previous experiences with prediction models, the predicted small-scale convective phenomena can undergo scores of lifecycles during the course of a single prediction period. The cloud evolution is thus highly nonlinear, or chaotic. Although there should be little predictability to an individual cloud (unless tied to a known topographical or land use feature), there is reason to believe that the coherent structures formed within the cloud field might be simulated skillfully. In this way, cloud-resolving models are more akin to “climate” models than to deterministic “prediction” models. Hence, the goals of such a model for the cloud prediction model are also similar to climate prediction models, i.e. to faithfully represent the statistical effect of the clouds on the large prediction scale. In addition, the predicted organization of the simulated cloud fields might be expected to have some integrity, at least to the extent that it is a function of a predictable large scale environment, as we believe. A series of numerical experiments aimed at developing a numerical scheme that faithfully simulates the observed coherent cloud structures is presented. The particular cloud structures studied were observed as part of the 1998 Lake-ICE field experiment designed to study cloud structures associated with lake-effect storms over Lake Michigan. Initial model results revealed an unsettling dependence of simulated structures on the numerical scheme exceeding the influence of the physical environment. Reasons for these dependencies are shown to be related to systematic biases of the truncation error. Corrections to these biases are shown to greatly improve the model’s performance in simulating the characteristics of simulated coherent structures.