Prediction of pollutant levels and their tendency is an important topic in environmental science today. To achieve such prediction task the use of neural networks, in particular the multi-layer per-ceplron, is regarded as a cost-effective technique superior to traditional statistical methods. In this paper neural networks were used to forecast the different long-term concentration levels for some of the well-known pollutants (sulphur dioxide, total suspended particulate, nitrogen dioxide, fluorides) in the urban area of Kamensk-Uralsky City. Concentration of a pollutant in atmospheric air as a function of parameters such as horizontal wind speed, trend of wind, temperature and concentration of this pollutant in time preceding the present is discussed. The neural network forecasts resulted as excellent, giving a generalization of problem with MSE about 1 % for all the models. The most successful model has shown very good performance for the 4 days forecasts. For local administrations and environmental protection institutions this methodology seems to be very useful.