Sequestering carbon (C) into stable soil pools has potential to mitigate increasing atmospheric carbon dioxide concentrations. Carbon accrues in grassland soil restored from cultivation, but the amount of physically protected C (here measured as microaggregate-within-macroaggregate C) and predominant mechanisms of accrual are not well understood. We modeled the rate of physically protected carbon accrued in three mesic temperate perennial restored grasslands from cross-continental regions using datasets with a wide range of restoration ages from northeast Kansas, USA; southeast Nebraska, USA; and northeast Free State, South Africa. Further, we investigated major controls on the amount of physically protected C in each site using structural equation modeling. Variables in the structural equation model were root biomass, root C:N ratio, soil structure (indicated by bulk density, percent of macroaggregates on a per whole soil mass basis, and percent of microaggregate-within-macroaggregates on a per macroaggregate mass basis), microbial composition (indicated by microbial biomass C, total phospholipid fatty acid [PLFA] biomass, and PLFA biomass of arbuscular mycorrhizae fungi [AMF] biomass), and microaggregate-within-macroaggregate C on a per whole soil mass basis. Across all sites, physically protected C accrued at a rate of 16 ± 5 g m−2 year−1. Data from South Africa fit an a priori metamodel developed for northeast KS that hypothesized physically protected C could be explained as a function of microbial composition, soil structure, root C:N ratio, and root biomass (listed in order of strength of direct effect on physically protected C). In contrast to the model-based hypothesis, root C:N ratio was the strongest influence (negative) on physically protected C in South Africa. The lesser effect of AMF on physically protected C in South Africa was consistent with lower AMF biomass in arid environments. The hypothesized model did not fit southeast Nebraska data possibly due to high (~ 30%) clay content. Overall, these results suggest that physically protected C in soil with moderate amounts of clay (more than 10% and less than 30%) can be predicted with knowledge of roots (biomass and C:N ratio), microbial biomass, and soil aggregation.