Abstract: Soil has the largest carbon pool in terrestrial ecosystems thus accurately assessing pool size and spatial distribution pattern is critical in soil ecology． Previous studies have shown that soil carbon density generally decreases from top to bottom along the soil profile, but mechanisms involved are not understood． Field measurements from 80 sampling sites in a subtropical evergreen broad-leaved forest in Chebaling, Guangdong were systematically evaluated for spatial variation in the 0–100 cm soil carbon density and decay coefficient β along the soil depth was calculated． Average soil carbon density in Chebaling forest was found to be 11.09 kg·m⁻², the mean decay coefficient (parameter β) was 0.961 1．Least subset square regression and structural equation modeling analyses revealed that β was predominantly determined by surface soil fine root biomass． Greater fine root biomass in the surface soil contributed to a smaller β, indicating greater carbon accumulation in surface soil． Other factors such as surface soil carbon-to-nitrogen ratio, total ambient radiation and aboveground biomass also contributed to β variations． Such factors totally could account for 66% of spatial variation in β． Simulated β values based on these factors combined with surface soil carbon density could effectively predict soil carbon density from 0 to 100 cm in depth． In conclusion, β and whole-profile soil carbon density could be well predicted using plant and surface soil characteristics． This approach could extrapolate abundant surface soil carbon data to estimate soil carbon stocks down to deep soil layer, to provide an efficient pathway for estimation of whole-profile soil carbon density in subtropical evergreen broad-leaved forests lacking deep soil samplings．