Abstract. It has been proposed that the large soil carbon (C) stocks of humid tropical forests result predominantly from C stabilization by reactive minerals, whereas oxygen (O2) limitation of decomposition has received much less attention. We examined the importance of these factors in explaining patterns of C stocks and turnover in the Luquillo Experimental Forest, Puerto Rico, using radiocarbon (14C) measurements of contemporary and archived samples. Samples from ridge, slope, and valley positions spanned three soil orders (Ultisol, Oxisol, Inceptisol) representative of humid tropical forests, and differed in texture, reactive metal content, O2 availability, and root biomass. Mineral-associated C comprised the large majority (87 ± 2%, n = 30) of total soil C. Turnover of most mineral-associated C (66 ± 2%) was rapid (11 to 26 years; mean and SE: 18 ± 3 years) in 25 of 30 soil samples across surface horizons (0–10 and 10–20 cm depths) and all topographic positions, independent of variation in reactive metal concentrations and clay content. Passive C with centennial–millennial turnover was typically much less abundant (34 ± 3%), even at 10–20 cm depths. Carbon turnover times and concentrations significantly increased with concentrations of reduced iron (Fe(II)) across all samples, suggesting that O2 availability may have limited the decomposition of mineral-associated C over decadal scales. Steady-state inputs of mineral-associated C were statistically similar among the three topographic positions, and could represent 10–25% of annual litter production. Observed trends in mineral-associated Δ14C over time could not be fit using the single-pool model used in many other studies, which generated contradictory relationships between turnover and Δ14C as compared with a more realistic two-pool model. The large C fluxes in surface and near-surface soils documented here are supported by findings from paired 14C studies in other types of ecosystems, and suggest that most mineral-associated C cycles relatively rapidly (decadal scales) across ecosystems that span a broad range of state factors.