In order to reduce uncertainties and enhance the knowledge of elevated releases atmospheric dispersion in a rural plain, the French Institute for Radioprotection and Nuclear Safety (IRSN), in collaboration with VEOLIA, carried out six weeks of experimental campaigns between November 2008 and July 2009 in the vicinity of an energy recycling unit. The atmospheric dispersion of the plume was studied by SF 6 tracer injection in the 40 m high stack. Maximal values of experimental Atmospheric Transfer Coefficient (ATCmax) and horizontal dispersion standard deviations ( h ) were compared to the results of the first generation Gaussian models, Doury and Briggs, and to the results of the last generation Gaussian models, ADMS 4.1 (Atmospheric Dispersion Modelling System). Several modelling parameterization for ADMS 4.1 computations were tested and revealed an overestimation of the h with the building option or with the integration of a surface roughness file. Consequently, ADMS 4.1 was used without model options. The Doury and Briggs models were combined with Holland formulation for the effective height calculation. In neutral atmospheric conditions and in summer unstable conditions (class A, B and C according to Pasquill classification), the ADMS 4.1 model is appropriate to estimate ATCmax value but overestimates in some cases. It was noticed that, during wintry periods and in class C atmospheric conditions, all of these models overestimated ATCmax for distances from the release point comprised between 0 and 2000 m. To estimate the atmospheric dispersion of an industrial release with a commercial model, as ADMS 4.1, without a prior comparison with an experimental data base dedicated to the studied site, can induce a poorly suitable modelling parameterization and leads to uncertainties difficult to quantify on the dispersion conditions.