Agricultural ecosystems are major sources of greenhouse gas (GHG) emissions, specifically nitrous oxide (N<sub>2</sub>O) and carbon dioxide (CO<sub>2</sub>). An important method of investigating GHG emissions in agricultural ecosystems is model simulation. Field measurements quantifying N<sub>2</sub>O and CO<sub>2</sub> fluxes were taken in a summer maize ecosystem in Zhangye City, Gansu Province, in northwestern China in 2010. Observed N<sub>2</sub>O and CO<sub>2</sub> fluxes were used for validating flux predictions by a DeNitrification-DeComposition (DNDC) model. Then sensitivity tests on the validated DNDC model were carried out on three variables: climatic factors, soil properties and agricultural management. Results indicated that: (1) the factors that N<sub>2</sub>O emissions were sensitive to included nitrogen fertilizer application rate, manure amendment and residue return rate; (2) CO<sub>2</sub> emission increased with increasing manure amendment, residue return rate and initial soil organic carbon (SOC); and (3) net global warming potential (GWP) increased with increasing N fertilizer application rate and decreased with manure amendment, residue return rate and precipitation increase. Simulation of the long-term impact on SOC, N<sub>2</sub>O and net GWP emissions over 100 yr of management led to the conclusion that increasing residue return rate is a more efficient method of mitigating GHG emission than increasing fertilizer N application rate in the study area.