Numerical models are widely used to predict wind flow and pollutant dispersion in industrial and urban landscapes. Many of them are already commercially available and are being applied by consulting engineers to complex real world problems. There is, however, some doubt about the evaluation of these models and the quality of predictions. This paper attempts to give a critical assessment of the current state of model evaluation, thereby focusing oil the most important step of the evaluation procedure: the comparison of model predictions with data. It goes without saying that the result of this process will only be as good as the reference data used. By using obstacle resolving CFD models it is shown that it is not an easy task to provide data that can be regarded as a reliable standard.