Human activities have clearly caused dramatic alterations of the terrestrial nitrogen cycle, and analyses of the extent and effects of such changes are now common in the scientific literature. However, any attempt to evaluate N cycling processes within ecosystems, as well as anthropogenic influences on the N cycle, requires an understanding of the magnitude of inputs via biological nitrogen fixation (BNF). Although there have been many studies addressing the microbiology, physiology, and magnitude of N fixation at local scales, there are very few estimates of BNF over large scales. We utilized >100 preexisting published estimates of BNF to generate biome- and global-level estimates of biological N fixation. We also used net primary productivity (NPP) and evapotranspiration (ET) estimates from the Century terrestrial ecosystem model to examine global relationships between these variables and BNF as well as to compare observed and Century-modeled BNF. Our data-based estimates showed a strong positive relationship between ecosystem ET and BNF, and our analyses suggest that while the model's simple relationships for BNF predict broad scale patterns, they do not capture much of the variability or magnitude of published rates. Patterns of BNF were also similar to patterns of ecosystem NPP. Our best estimate of potential nitrogen fixation by natural ecosystem is ~195 TgNyr-1, with a range of 100--290 TgNyr-1. Although these estimates do not account for the decrease in natural N fixation due to cultivation, this would not dramatically alter our estimate, as the greatest reductions in area have occurred in systems characterized by relatively low rates of N fixation (e.g., grasslands). Although our estimate of BNF in natural ecosystems is similar to previously published estimates of terrestrial BNF, we believe that this study provides a more documented, constrained estimate of this important flux. ¿ 1999 American Geophysical Union