Abstract. Confidence in regional estimates of N20 emissions used in national greenhouse gas inventories could be improved by using mathematical models of the biological and physical processes by which these emissions are known to be controlled. However these models must first be rigorously tested against field measurements of N20 fluxes under well documented site conditions. Spring thaw is an active period of N20 emission in northern ecosystems and thus presents conditions well suited to model testing. The mathematical model ecosys, in which the biological and physical processes that control N20 emissions are explicitly represented, was tested against N20 and CO2 fluxes measured continuously during winter and spring thaw using gradient and eddy covariance techniques. In the model, ice formation at the soil surface constrained soilatmosphere gas exchange during the winter, causing low soil 02 concentrations and consequent accumulation of denitrification products in the soil profile. The removal of this constraint to gas exchange during spring thaw caused episodic emissions of N20 and CO2, the timing and intensities of which were similar to those measured in the field. Temporal variation in these emissions, both simulated and measured, was high, with those of N20 ranging from near zero to as much as 0.8 mg N m -2 h -• within a few hours. Such variation should be accounted for in ecosystem models used for temporal integration of N20 fluxes when making long-term estimates of N20 emissions.