In fallow-based agriculture (FBA), the ratio in which the annual crop land is divided between cropping and fallowing, the land use factor, is determined by the trade-off between land idleness and cost-free fertilization. Previous research on FBA dynamics, conducted by economists, has implicitly assumed that farmers have perfect knowledge of the relation between the land use factor and yield. This hypothesis is relaxed on the paper, being replaced by the assumption that agents tend to deviate from the so-called optimal land use factor in a magnitude that decreases with the knowledge about FBA. In such framework, learning about FBA matters. Field evidence reveals that learning occurs gradually, through experimenting with alternative fallow durations and, thus, testing the limits of FBA systemic resilience. What might explain why some authors sustain that Amazonian farmers tend to practice fallow durations that are inconsistent with a sustainable level of yield, creating poverty-traps where soil quality and agricultural tend to decay progressively, and consequently, farmers' standard of living. Whether this is led by credit or capital constraints or by the learning process through which the sustainable fallow duration is sought is an open question that has not yet being submitted to empirical refutation. This paper aims to fill this gap, by bringing insights from the literature on the economics of learning to understand how Brazilian Amazon farmers manage fallow. An empirical model is proposed to test the hypothesis that proxies for knowledge are statistically significant predictors of the crosssectional variation of the effective land use factor. The hypothesis is not refuted by farm-level data, gathered from a survey conducted in three municipalities of Pará state, Brazilian Amazon, even controlling by factors that determine the optimal level of fallow duration. What suggests that a policy that aims to eradicate FBA, or to replace it for some "green-revolution" surro