To help design advisory approaches tailored to farmers' needs, we propose to highlight the different learning processes that lead to more sustainable farms. To this end, we have defined the concept of 'learning regime' as the set of mechanisms that are triggers for and lead to the acquisition of new knowledge and skill s, allowing the head of the farm to improve his production and management methods. This concept has been tested in the cotton-growing basin of Bobo-Dioulasso in 30 farms considered innovative in five technical areas related to environmental sustainability. The results show that there are four different types of regimes (transformer-observer, reactive-networker, optimizer-self-reliant and imitator-dependent). These results call into question the assumption of homogeneity of farmers' capacities to change their routines to acquire new skills and know-how. However, give n that the nature of technical changes implemented is heavily influenced by the informational context in which the producer operates, supporting the learning of sustainability calls into question, in particular, the ability to empower farmers' vis-à-vis this context. While this work's contributions are currently mainly conceptual and methodological, it opens up new perspectives to improve processes and tools to support the emergence of a more sustainable agriculture .