Accurate knowledge of carbon (C) content in live wood is essential for quantifying tropical forest C stocks, yet generic assumptions (such as biomass consisting of 50% carbon on a weight/weight basis) remain widely used despite being supported by little chemical analysis. Empirical data from stem cores of 59 Panamanian rainforest tree species demonstrate that wood C content is highly variable among co-occurring species, with an average (47.4±2.51% S.D.) significantly lower than widely assumed values. Prior published values have neglected to account for volatile C content of tropical woods. By comparing freeze- and oven-dried wood samples, we show that volatile C is non-negligible, and excluding the volatile fraction underestimates wood C content by 2.48±1.28% (S.D.) on average. Wood C content varied substantially among species (from 41.9-51.6%), but was neither strongly phylogenetically conserved, nor correlated to ecological (i.e. wood density, maximum tree height) or demographic traits (i.e. relative growth rate, mortality rate). Overall, assuming generic C fractions in tropical wood overestimates forest C stocks by ∼3.3-5.3%, a non-trivial margin of error leading to overestimates of 4.1-6.8 Mg C ha(-1) in a 50-ha forest dynamics plot on Barro Colorado Island, Panama. In addition to addressing other sources of error in tropical forest C accounting, such as uncertainties in allometric models and belowground biomass, compilation and use of species-specific C fractions for tropical tree species would substantially improve both local and global estimates of terrestrial C stocks and fluxes.