Dust dispersion in an urban area becomes a major concern in several fields: global safety, pollution tracking, accidental release of highly active substances in powder form.In Switzerland, this last point became a relevant scenario of major accident in the middle of 2015. Every company producing or working with highly active substances in powder form (e.g. pharmaceutical and chemical firms) must thus assess the consequences of the dispersion of this kind of product and based on the results of the simulations, the authorities grant their approval for the production of this powder.The main background of dust dispersion modeling relies on heavy gas dispersion modeling. Indeed, air loaded with dust has an apparent density higher than the ambient one and behaves globally as a heavy gas. But other phenomena such as sedimentation, agglomeration have to be considered. Furthermore, in complex configurations such as urban areas, the accuracy of heavy gas models is low.This paper aims to evaluate the efficiency of an Artificial Neural Networks model to predict the dust dispersion in an urban area. Dust concentration data were collected at different places in a city.The wind velocity, direction, and atmospheric temperature were measured at nearest Meteoswiss station. This one year long data acquisition is thus a very rare data set that can be really useful to calibrate dust dispersion models in such areas.Results about the comparison between the experimental concentrations found and the results of Artificial Neural Networks based model of dust dispersion are presented. The results are discussed to explain the trends of the experimental values and the variation of accuracy of the tested models.