1. Regional above-ground biomass estimates for tropical moist forests remain highly inaccurate mostly because they are based on extrapolations from a few plots scattered across a limited range of soils and other environmental conditions. When such conditions impact biomass, the estimation is biased. The effect of soil types on biomass has especially yielded controversial results. 2. We investigated the relationship between above-ground biomass and soil type in undisturbed moist forests in the Central African Republic. We tested the effects of soil texture, as a surrogate for soil resources availability and physical constraints (soil depth and hydromorphy) on biomass. Forest inventory data were collected for trees >= 20 cm stem diameter in 2754 0.5 ha plots scattered over 4888 km(2). The plots contained 224 taxons, of which 209 were identified to species. Soil types were characterized from a 1:1 000 000 scale soil map. Species-specific values for wood density were extracted from the CIRAD's data base of wood technological properties. 3. We found that basal area and biomass differ in their responses to soil type, ranging from 17.8 m(2) ha(-1) (217.5 t ha(-1)) to 22.3 m(2) ha(-1) (273.3 t ha(-1)). While shallow and hydromorphic soils support forests with both low stem basal area and low biomass, forests on deep resource-poor soils are typically low in basal area but as high in biomass as forests on deep resource-rich soils. We demonstrated that the environmental filtering of slow growing dense-wooded species on resource-poor soils compensates for the low basal area, and we discuss whether this filtering effect is due to low fertility or to low water reserve. 4. Synthesis. We showed that soil physical conditions constrained the amount of biomass stored in tropical moist forests. Contrary to previous reports, our results suggest that biomass is similar on resource-poor and resource-rich soils. This finding highlights both the importance of taking into account soil characteristics and species wood density when trying to predict regional patterns of biomass. Our findings have implications for the evaluation of biomass stocks in tropical forests, in the context of the international negotiations on climate change.