Few studies explicitly assess the temporal and spatial dynamics of agroforestry adoption occurring beyond the project cycle. Where ex-post evaluations are published, abandonment of introduced agroforestry after project cessation is often reported. This paper presents an analysis of agroforestry adoption in a poor, peri-urban village in semi-arid south India, where 97 % of initial adopters had retained their plots six to eight years after implementation. The intervention was facilitated by BAIF, an Indian non-governmental organisation specialising in natural resource management. The complex technological package promoted was known as ‘wadi’ and comprised fruit trees planted in crop fields, with a boundary of multi-purpose trees and integrated soil and water conservation measures. Sixty four agroforestry plots belonging to 43 households were surveyed in 2010/11 and interviews were held with both adopting and non-adopting farmers. Beyond retention, a quarter of adopters had expanded the practice on to additional areas of land and some diffusion to initially non-adopting farmers had also occurred. Adopters were found to have modified the practice to suit their own objectives, capabilities and constraints, highlighting that adoption is more than a simple binary choice. The study demonstrates the importance of external support for adoption of agroforestry. The intervention was not, however, especially pro-poor with adoption occurring disproportionately among relatively wealthier households with larger landholdings. Where poorer households adopted, this tended to occur later. Participation was entirely voluntary and, by 2011, conversion of suitable farmland to agroforestry had reached 18 %; while beneficial to individual adopters, this patchy coverage arguably limits the potential for enhanced ecosystem service provision at landscape-scale.