Palms (family Arecaceae) are abundant in Amazonian forests, but the allometry of these monocotyledonous plants remains poorly quantified. Woody palm biomass is most commonly estimated with dicotyledonous tree models, which leaves substantial uncertainty as to their true biomass and productivity. We developed the first extensive dataset of directly-measured arborescent palm biomass: 136 individuals from nine species in terra firme and wetland forests - Astrocaryum murumuru, Attalea phalerata, Bactris gasipaes, Euterpe precatoria, Iriartea deltoidea, Mauritia flexuosa, Mauritiella aculeata, Oenocarpus bataua, and Socratea exorrhiza. We created single species (n= 8-21) and family-level (n= 97-106) allometric equations, using diameter, stem height, total height, and stem dry mass fraction, to estimate (i) total aboveground biomass for all species, (ii) belowground biomass for the two wetland species (Mauritia and Mauritiella), and (iii) leaf mass for all species. These new palm models were then applied to nine 1-ha plots in the southwestern Amazon (Tambopata) to calculate the impact on forest biomass estimates once palm mass is estimated with palm-specific models, rather than from models created for dicot trees. We found that stem height was the best predictor variable for arborescent palm biomass, but the relationship between stem height and biomass differed among species. Most species showed weak biomass-diameter relationships, but a significant relationship could be identified across all species. The new palm models were better estimators of palm mass than existing dicot models. Using our species-level models increased estimates of palm biomass at our study site by 14-27%, compared to using recently published pantropical biomass models for trees. In other forests, the effect of using these palm equations on biomass estimates will depend on palm sizes, abundance, and species composition. © 2013 Elsevier B.V.