Abstract Experiments were carried out to check the validity of the method used to calculate gas fluxes in the wind-tunnel technique. This method assumes that the airflow is completely mixed downwind of the emission plot. This hypothesis was tested using a homogeneously distributed CO2source installed on the ground under the tunnel cover. Carbon dioxide was used instead of ammonia to allow measurement of its concentration fluctuation. The measured vertical profiles of wind speed and concentration were non-uniform in the measurement section, showing that the flow was far from being completely mixed, and leading to a recovery rate ranging from 77 to 87%. Using the measured concentration and wind profiles, the efficiency of several sampling systems was numerically simulated. Using this approach, the conventional sampling system led to an underestimation of the actual flux of up to 11%, while the non-uniformity of the wind profiles was responsible for an overestimation of 3%. The turbulent component of the horizontal flux was found to be less than 0·5% of the mean component. This study suggests that an optimal sampling system could decrease the error due to sampling to a small percentage, while a flow modifier could halve the error due to the wind profile.