This paper outlines a mesoscale forecast system that could he implemented within a few years in spite of relatively sparse direct observations. Methods are discussed of using satellite information on large mesoscale features to initiate numerical models. The models develop further mesoscale structures through the influence of mesoscale geographic features and organized convective systems. The output of the numerical model serves as the physical foundation upon which the latest detailed satellite data can he interpreted. Although many of the techniques described are not off-the-shelf items today, they are entirely feasible. It is important that the components of the forecast system be developed in parallel, rather than in series, if the system is to be completed within five years. The components include: polar-orbiting satellites for high latitudes; geosynchronous satellites for low latitudes; a mesoclimatological data base largely from satellite data; a mesoscale numerical prediction model with lateral boundary data supplied from a conventional large-scale model, and a variety of simple models and empirical schemes for treating special mesoscale phenomena. A review of current activities in mesometeorology provides substantial evidence that the revolution in large-scale weather prediction in the past decade will be followed by a similar revolution on the mesoscale in the next five years.