Although agroforestry is recognised as a means to stabilise farm income, little work has been done to differentiate farmers based on their incomes, their capacity to diversify crops, and the economic outcomes of agroforestry adoption. By distinguishing between high- and low-income farmers, this paper examines how agricultural training affects agroforestry promotion, and evaluates its relevance for the poor, the extent of its adoption, and its economic consequences. We found that although training generally increased participants’ knowledge of agroforestry, it increased crop diversity only for low-income participants. We also detected the presence of information spillovers from participants to non-participants, which may increase crop diversity among non-participants and consequently reduce the measurable impact of training. When income heterogeneity is considered, we found that the low-income participants benefited more from increasing their incomes and expanding their social networks than the higher-income participants. We also found that agroforestry adoption helped to reduce income volatility.