This paper aims to modify some ideas about frontier development in the Amazon region. Modelling the development of frontiers first colonised by family farmers, many authors show that poor soil, colonisation policies, and economic logic leads to rapid development of extensive cattle ranching and thus the need for more pasture land, which leads in turn to deforestation, migration of family farmers and concentration of the land in the hands of a few landowners. This paper addresses the conditions for the sustainability of family agriculture in an Amazonian frontier, by considering the diversity of its biophysical environments and the socioeconomic characteristics and skills of the populations living from its natural resources. Our aim is to look for experiences that cannot be identified by a global analysis but that could, with support, constitute other ways to imagine use of the frontier lands. These different uses are what we call discrete innovations. We show that the soil fertility trends result, above all, from the logic of farmers who decide not to take the diversity of the environment into account in their land management. This diversity must be considered in any future analysis of the sustainability of agricultural activity, and we try to find ways to ensure both. The history of family farmers is important in explaining why they have not tried to restore the soil. Although their migrations and the rapid development of cattle ranching can be seen as due to the political and economical context, we argue that this adaptation must be understood as a consequence of the family reproduction strategies of the Brazilian peasants. At the same time, we find that their technical knowledge depends on their personal history and that the diversity of their skills and practices affects the management of the vegetation cover. We also identify innovations conceived by farmers who sought to use the diversity of the environment to make their lives and living more stable; these are innovations that can be generalised.