We used evolutionary programming to model innate migratory pathways of wildebeest in the Serengeti Mara Ecosystem, Tanzania and Kenya. Wildebeest annually move from the southern short-grass plains of the Serengeti to the northern woodlands of the Mara. We used satellite images to create 12 average monthly and 180 10-day surfaces from 1998 to 2003 of percentage rainfall and new vegetation. The surfaces were combined in five additive and three multiplicative models, with the weightings on rainfall and new vegetation from 0% to 100%. Modeled wildebeest were first assigned random migration pathways. In simulated generations, animals best able to access rainfall and vegetation were retained, and they produced offspring with similar migratory pathways. Modeling proceeded until the best pathway was stable. In a learning phase, modeling continued with the ten-day images in the objective function. The additive model, influenced 25% by rainfall and 75% by vegetation growth, yielded the best agreement, with a multi-resolution comparison to observed densities yielding 76.8% of blocks in agreement (kappa = 0.32). Agreement was best for dry season and early wet season (kappa = 0.22-0.57), and poorest for the late wet season (0.04). The model suggests that new forage growth is a dominant correlate of wildebeest migration.