Agroforestry systems are characterized by a high complexity between vegetation components and niche partitioning. In a crop-grass-tree agroforestry system, rape, willow, and grasses were in situ pulse labeled separately with 13CO2 for 6 h, and 13C was traced in shoots, roots, topsoil (0–15 cm) and subsoil (15–30 cm), microbial biomass carbon (C), and dissolved organic C, as well as respiration losses (CO2) up to 28 days after labeling to investigate the effects of vegetation components on C allocation belowground. 13C recovery in roots after 28 days was 7.0% of total assimilated C for grassland, which was 3.5- and 5.2-fold higher than that for rape and willow, respectively. The larger C allocation belowground in grassland was ascribed to its higher root/shoot ratio compared to willow and rape. Grassland facilitated higher accumulation of root-derived C in soil compared to rape (9.2% of recovered 13C) and compared to willow (1.6% of 13C). Willow retained more photosynthetic C aboveground and less was allocated to roots compared to rape. Although the C allocated to the top 15-cm soil was similar between willow and rape, willow facilitated C allocation in deeper soil compared to rape (0.6% vs. 0.2%). This could be explained by the lower microbial activity and subsequent weaker decomposition of rhizodeposits in 15–30-cm depth under willow. The net belowground C inputs in grassland, willow, and rape were 0.53, 0.06, and 0.10 g C m−2 month−1 of vegetation period, including rhizodeposition of 0.24, 0.05, and 0.04 g C m−2 month−1, respectively. Overall, integrating trees and grassland within cropland facilitates higher root-derived C input into soil, thus contributing to the soil C sequestration in agroforestry systems.