NO soil emissions are directly influenced by soil environmental (temperature, humidity), chemical (pH, N content, C content…) and physical (soil content, texture) variables. All these parameters exert linear or non linear influences that fluctuate in threshold and intensity between sites. Because of the lack of field experiments and to the high variability in time (diurnal and seasonal cycle) and space (regions, soil and vegetation type) of the environmental parameters influencing NO emissions, estimates of NO emissions worldwide still remain highly uncertain. In this study we developed nonlinear regressions to describe NO flux emission from soil in dependency with relevant environmental parameters for a forest site in a temperate region (Höglwald, South Germany, 1994–1997) using an Artificial Neural Network (ANN). The resulting algorithm links NO fluxes with air, surface and depth temperatures, surface WFPS (Water Field Pore Space) and humus pH. All these parameters were evaluated and selected as relevant and non redundant. Network performances are evaluated for different numbers of hidden neurons. Resulting equations linking NO fluxes from soils and variables are obtained, and show to perform well with measurements (R2 = 0.81). Average NO fluxes values of 14.6 gN ha−1 d−1 are obtained for calculated and measured fluxes. In a second part, 2002–2003 NO soil fluxes are estimated from the ANN equation obtained from 1994–1997 flux measurements performed at the same site. Overall, simulated results give a good estimation of NO fluxes, with a mean value of 15.3 gN ha d−1 close to the 21.7 gN ha d−1 measured mean for the 2002–2003 period. ANN algorithm gives also a good representation of low frequency (seasonal) variations. On the basis of our results, we suggest that ANN is a good alternative between detailed biogeochemical models and large scale models, and may be the appropriate tool for estimating NO emissions at a regional scale.