The assessment of forest biomass has important consequences for climate-related policies, including conservation and activities related to deforestation and forest degradation. Tree and forest biomass are most efficiently assessed by a combination of remote-sensing, ground-based measurements and allometric models to predict tree and forest characteristics such as tree volume and biomass. The accuracy of these allometric models and therefore of biomass estimates can be improved greatly if wood density is used as an input variable. It is the most commonly used species functional trait in multispecies biomass models, which are frequently used in tropical ecosystems. However, wood density is difficult to measure accurately and thus is not included in large-scale forest inventories. This issue can be bypassed by using species average values, but often wood density values are only available at coarse taxonomic and/or spatial scales. Applying regional multi-species averages reintroduces to the models the potential bias originally removed by the use of wood density values. Therefore more wood density data is needed to cover the diversity of species and environments. This paper presents the work carried out to complement current initiatives aiming to collect, harmonize and share existing wood density data stored in national research institutes across the tropics. The database compiles raw data and calculated averages, with 70 information fields including GPS coordinates, measurement methods and sample size, when available, for a wide range of taxonomic and environmental values. Where precision in both taxonomic and environmental scales remains difficult to meet, the trade-off between the two can be better understood as the database includes both very accurate data and coarse averages. This database will improve forest biomass estimates, towards increasing the potential of climate change mitigation initiatives in the forestry sector.