The decentralization of natural resource management provides an opportunity for communities to increase their participation in related decision making. Research should propose adapted methodologies enabling the numerous stakeholders of these complex socioecological settings to define their problems and identify agreed-on solutions. This article presents a companion modeling (ComMod) experiment combining role-playing games and multiagent systems conducted in a community in northern Thailand to support collective learning for adaptive land management. Researchers and local stakeholders collectively built a representation of the situation and used it as a platform to explore scenarios. This ComMod process initially addressed a soil erosion problem. The participants identified the expansion of perennial crops as a promising solution but also raised the problem of the unequal ability among villagers to invest in such crops. The researchers flexibly adapted the simulation tools to the emerging matter. The authors assess the learning effects of this experiment and identify two favoring factors: the increasing participation of local stakeholders and a flexible and adaptive modeling process suited to learning, which by nature is an evolving process. But to ensure sustainable impacts for the communities, stronger links with higher institutional levels are needed. (Contains 6 figures, 1 table and 2 notes.)