Alternating degradation and restoration phases of soil quality, as is common in crop-fallow systems, can be avoided if the restorative elements of trees and forests can be integrated into productive agroforestry systems. However, evidence for the hypothesis of ‘internal restoration’ in agroforestry is patchy and the effectiveness may depend on local context. We investigated to what extent cocoa (Theobroma cacao, L.) agroforestry can recover soil structure and infiltration in comparison to monoculture systems across the Konaweha Watershed, Southeast Sulawesi. We compared soil organic carbon, fine root length and weight, soil aggregate stability, macroporosity and infiltration from three soil layers at five land use systems: i.e. degraded forests, 9–14 years old of complex-cocoa agroforestry, simple-cocoa agroforestry, monoculture cocoa and 1–4 years old annual food crops, all with three replications. In general, roots were concentrated in the upper 40 cm of soil depth, contained of 70% and 86% of total fine root length and weight. Compared to simple agroforestry and cocoa monoculture, complex agroforestry had greater root length and weight in the topsoil, even though it attained only half the values found in degraded forests. Higher root density was positively correlated to soil organic carbon. In upper soil layers, complex agroforestry had slightly higher soil aggregate stability compared to other agricultural systems. However, no significant difference was found in deeper layers. Complex agroforestry had higher soil macroporosity than other agricultural systems, but not sufficient to mimic forests. Degraded forests had two times faster steady-state soil infiltration than agricultural systems tested (13.2 cm h−1 and 6 cm h−1, respectively), relevant during peak rainfall events. Compared to other agricultural systems, complex agroforestry improves soil structure of degraded soil resulting from forest conversion. However, a considerable gap remains with forest soil conditions.