Forests and savannas are the major ecotypes in humid tropical regions. Under present climatic conditions, forest is in a phase of natural expansion over savanna, but traditional human activities, especially fires, have strongly influenced the succession. We here present a new model, FORSAT, dedicated to the forest–savanna mosaic on a landscape scale and based on stochastic modelling of key processes (fire and succession cycle) and consistent with common field data. The model is validated by comparison between the qualitative emergent behaviour of the model and results of biogeographical field studies. Three types of forest succession are shown: progression of the forest edge, formation and coalescence of clumps in savanna and global afforestation of savanna. The parameters (frequency of savanna fires, climate and soil fertility) appear to have comparable effects and there is a sharp threshold between a forest edge progression scenario and the cluster formation one. Moreover, pioneer seed dispersal pattern and recruitment are determinant: peaked curves near a seed source and far dispersal combine to increase the fitness of the pioneers.