This paper presents computational simulations of atmospheric dispersion experiments conducted around isolated obstacles in a wind tunnel. The tool used for the simulations is the computational fluid dynamics (CFD) code ADREA-HF, which was especially developed for the calculation of flow and dispersion of positively or negatively buoyant gases in complicated geometries. The wind tunnel experiments simulated involve a cube normal to the flow, a taller rectangular building—comprising of two stacked cubes and a right circular cylinder. Three different gas source locations are examined: two cube heights upwind, at the upwind face and at the downwind face of each obstacle. The experimental data in all cases consist of mean concentrations and concentration fluctuations downwind of the obstacle. In the first part of the study, a computational assessment is performed to examine the influence of factors such as turbulence model, grid resolution, boundary conditions and numerical scheme, on the results of the CFD model in the case of dispersion around an isolated cube. Following this assessment the model parameters are optimized. The model is then used so that computed flow fields and concentration patterns around the three obstacles and for the three different source positions are inter-compared and analysed. Along-wind profiles of computed and measured concentrations and concentration standard deviations have been compared to examine the differences between simulations and wind tunnel experiments. Finally, a statistical performance evaluation of the model is carried out by comparing computed and experimental concentrations and concentration fluctuations. In most cases there is a good level of agreement between calculated and measured quantities, while the model has a general tendency to over-predict concentration fluctuations. In conclusion, the wind tunnel data together with the detailed spatial results that the CFD model produces, give the opportunity to study in detail the flow fields and concentration patterns and to reveal the different behaviours associated with the different obstacle shapes and gas source locations.