An approach to simulating land-cover change based on pairs of classified images is presented. The method conditions the simulations on three sources of information: an initial land-cover map, maps of the probabilities of each possible class transition, and a description of the spatial patterns of changes (e.g., semivariograms). The method can produce multiple simulated land-cover maps that honor each of these sources of information. The approach is demonstrated for data on forest-cover change near Traverse City, Michigan. The discussion describes extensions to the method and an approach to generating future land-cover scenarios based on socioeconomic information.