Five atmospheric dispersion experiments, all assigned Pasquill stability class D, were performed at Risø National Laboratory. The tracer sulphurhexafluoride was released at a height of 60 m from the Risø meteorological tower, situated on a peninsula in the Roskilde Fjord, Denmark, and was sampled at ground level about 5 km downwind. The experiments were all performed with westerly winds. The dispersion was influenced by the internal boundary layer that develops downwind from the inhomogeneity between the Roskilde Fjord and the Risø site. During the experiments, turbulent wind velocity fluctuations were measured at the height of release with a three-dimensional ultrasonic anemometer. The lateral dispersion parameter, σy, was estimated according to three methods based on quantitative descriptions of the flow, and compared to the value of σy, derived from the measured tracer concentrations. The methods described in Hay and Pasquill (1959), and Draxler (1976) were found to work well in predicting σy in the actual experiments. Equivalent comparisons of the vertical dispersion parameter, σz, were troublesome, as the sampling unit set-up only allowed for an indirect estimate of σz, based on continuity considerations. But the method described in Draxler (1976) was found to predict the indirect measured σz reasonably well when taking the effect of the developing internal boundary layer into account. The ability of the various methods to predict the maximum ground level tracer concentration was investigated. The agreement between the estimated maximum ground level concentrations derived from Draxler (1976) and the concentrations actually measured was found to be better in most cases than the absolute calibration uncertainty in the tracer analysis.