We report on an ecosystem modeling approach that integrates global satellite, climate, vegetation, and soil data sets to (1) examine conceptual controls on nitrogen trace gas (NO, N2O, and N2) emissions from soils and (2) identify weaknesses in our bases of knowledge and data for these fluxes. Nitrous and nitric oxide emissions from well-drained soils were estimated by using an expanded version of the Carnegie-Ames-Stanford (CASA) Biosphere model, a coupled ecosystem production and soil carbon-nitrogen model on a 1¿ global grid. We estimate monthly production of NO, N2O, and N2 based on predicted rates of gross N mineralization, together with an index of transient water-filled pore space in soils. Analyses of model performance along selected climate gradients support the hypothesis that low temperature restricts predicted N mineralization and trace gas emission rates in moist northern temperate and boreal forest ecosystems, whereas in tropical zones, seasonal patterns in N mineralization result in emission peaks for N2O that coincide with wetting and high soil moisture content. The model predicts the annual N2O:NO flux ratio at a mean value of 1.2 in wet tropical forests, decreasing to around 0.6 in the seasonally dry savannas. Global emission estimates at the soil surface are 6.1 Tg N and 9.7 Tg N yr-1 for N2O and NO, respectively. Tropical dry forests and savannas are identified by using this formulation as important source areas for nitrogen trace gas emissions. Because humans continue to alter these ecosystems extensively for agricultural uses, our results suggest that more study is needed in seasonally dry ecosystems of the tropics in order to understand the global impacts of land use change on soil sources for N2O and NO. ¿ American Geophysical Union 1996