Inverse-dispersion techniques allow inference of a gas emission rate Q from measured air concentration. In "ideal surface layer problems," where Monin-Obukhov similarity theory (MOST) describes the winds transporting the gas, the application of the technique can be straightforward. This study examines the accuracy of an ideal MOST-based inference, but in a nonideal setting. From a 6 m ϫ 6 m synthetic area source surrounded by a 20 m ϫ 20 m square border of a windbreak fence (1.25 m tall), Q is estimated. Open-path lasers gave line-averaged concentration C L at positions downwind of the source, and an idealized backward Lagrangian stochastic (bLS) dispersion model was used to infer Q bLS . Despite the disturbance of the mean wind and turbulence caused by the fence, the Q bLS estimates were accurate when ambient winds (measured upwind of the plot) were assumed in the bLS model. In the worst cases, with C L measured adjacent to a plot fence, Q bLS overestimated Q by an average of 50%. However, if these near-fence locations are eliminated, Q bLS averaged within 2% of the true Q over 61 fifteen-minute observations (with a standard deviation Q/Q ϭ 0.20). Poorer accuracy occurred when in-plot wind measurements were used in the bLS model. The results show that when an inverse-dispersion technique is applied to disturbed flows without accounting for the disturbance, the outcome may still be of acceptable accuracy if judgment is applied in the placement of the concentration detector.