This study is focused on the development of a Bayesian network for air quality assessment and aims at offering a pragmatic and scientifically credible approach to modelling complex systems where substantial uncertainties exist. In particular, the main object is the prediction of the occurrence of suitable conditions for the stagnation of pollutants in a given area. The analytical modeling of the network provides a set of independent nodes, represented by the outputs of a forecasting meteorological Limited Area Model, from which descend the conditions for the stagnation of pollutants in different areas of the city (through measurements of the heuristic pollutant from monitoring stations) and finally the global conditions. The urban area of Genoa (Italy) was selected in order to test the actual capability of the model prototype. Network training was performed by means of historical data resulting from significant statistical series of the past years by the air quality-monitoring network. The system used for data assimilation, construction and network learning is completely based on an open source statistical processing software.