Questions: We studied the importance of soil properties and neighbouring forest cover in affecting plant community biomass and assembly during the tropical forest restoration process. We also investigated how compositional responses depended on traits expected to influence individual success. Location: Forest restoration sites across anthropogenic grasslands in mixed use agricultural watersheds, eastern São Paulo state, Brazil. Methods: We identified and measured all woody individuals (DBH ≥ 5 cm) in four 200-m2 plots per site. Then we translated these measurements into above-ground biomass (AGB), and related AGB variability to neighbouring forest cover, soil texture and chemical fertility using mixed effect models. We assessed the effect of these predictors on different species groups, arranged according to variation in wood density, tree height or habitat selectivity, through multivariate abundance models. Results: AGB ranged between 0 and 104.7 t/ha (median of 10.4 t/ha), with high variation within, as well as between, watersheds. Sand percentage, forest cover and the interaction between soil nutrient concentrations and sand percentage were good predictors of measured AGB. The most parsimonious model projected a seven growing seasons AGB recovery of 70.90 t/ha, when a site is on fertile soils with 10% sand and surrounded by forest cover of 50%. In contrast, only 5.24 t/ha is predicted on acidic-poor soils with 67% sand and 0% forest cover. Increasing forest cover favoured smaller trees and habitat generalists while increasing sand percentage inhibited taller species and forest specialists. Sand percentage constrained softwoods in fertile soils. Conclusion: Our results confirm that the likelihood of restoration to pre-disturbance conditions is constrained in contexts of higher degradation, such as when agricultural use adversely affects soil properties and/or motivates extreme deforestation. Lower AGB found on sandy soils suggests that forest recovery is sensitive to local drought intensification. Given regional projections for extended dry seasons, restoration approaches could consider targeting alternative reference states, rather than historical/undisturbed ones, under highly altered environments, while aiming to improve soil and microclimate conditions to allow moist tropical forest recovery where feasible.