Agroforestry is widely purported to improve the livelihoods of smallholder farmers, rehabilitate degraded landscapes, and enhance the provisioning of critical ecosystem services, including carbon sequestration. Yet, the evidence base supporting these longer-term impacts is weak. Using a quasi-experimental evaluation design based on a theory-based and mixed methods framework, we investigated both the downstream and intermediate effects of a nine year effort led by Vi Agroforestry (herein Vi), a Swedish non-governmental organization (NGO), to promote agroforestry in large sections of Bungoma and Kakamega counties in western Kenya. In particular, we compared two sets of households against various outcome measures along the causal chain: those belonging to (a) 226 pre-existing farmer groups operating in 60 targeted programme villages; and (b) 206 pre-existing farmer groups operating in 61 geospatially matched comparison (non-programme) villages. To further counter selection bias, we combined several econometric analytical methods, including two-stage least squares regression (2SLS), with difference-in-differences estimation. In addition, to triangulate key findings and interrogate impact pathways, unforeseen outcomes, and unexpected quantitative results, we carried out semi-structured in-depth interviews with a sub-sample of 40 purposively selected programme participants. We also applied process tracing to investigate the linkages between Vi’s programme and previous agroforestry research carried out by the World Agroforestry Centre (ICRAF). We found these research-to-programme linkages to be strong, and that greater—albeit variable and generally modest—programme exposure and agroforestry uptake took place among the farmer groups Vi targeted. Similarly, significant, yet again variable, effects were also identified for agroforestry product income, fuelwood access, and milk yields among dairy farmers. Soil organic carbon (estimated through remote sensing) increased at a higher rate overall in the sampled farm plots of the programme villages, but, ironically, so too did soil erosion. Finally, we found limited evidence that the programme significantly bolstered food security, shock resilience, and education progression and spending. However, we identified statistically significant—although, again, modest—programme effects for our asset and consumption expenditure measures (which includes our primary outcome variable), particularly among households represented by female programme participants.