<p>Agroforestry has been promoted in northwestern Vietnam as a sustainable land management practice that offers multiple benefits, including livelihood enhancement, income diversification, adaptation to climate variability and reduced soil erosion on steep slopes. However, research on the multi-stage adoption process of agroforestry and the complex web of factors influencing farmers' decisions remains scarce. Our study addresses this gap by investigating the adoption process of fruit tree-based agroforestry practices in the upland farming systems of Son La, a northwestern province of Vietnam. Using a Bayesian Network model, we analyzed the combined effects of socio-economic, household and practice-specific factors that shape farmers’ attitudes and intentions throughout the adoption process, from initial decision-making to sustained use. The Bayesian Network was developed using both qualitative and quantitative inputs from 124 farmer interviews and workshops involving domain experts and farmers. Additionally, through an adoption pathway analysis of 124 interviewed farmers, we explored the diverse trajectories farmers follow to integrate agroforestry into their systems, highlighting the adaptive and non-linear nature of the adoption process. Our findings suggest that effective agroforestry promotion requires coordinated, holistic strategies and sustained support rather than single-point interventions or binary adoption models. This study contributes to the understanding of agroforestry adoption as a dynamic, context-sensitive process and provides insights for designing support mechanisms that align with farmers’ evolving needs and decision-making pathways.</p>