The intellectual system of forecasting of pollution of an atmosphere is offered. The structure of construction of intellectual system of forecasting is considered. Algorithms of functioning of the basic subsystems are presented. Results of experiments are resulted. In many problems of data processing including problems of forecasting, use of traditional algorithms is complicated in a kind of that data can be incomplete and-or inconsistent. Therefore it is expedient to apply to the decision of such problems neural networks and evolutionary technologies. Thus a perspective direction is use of hybrid systems and hierarchical neural networks. The intellectual system of forecasting of pollution of an atmosphere is offered. The system is based on hierarchical association of two neural networks: Коhonen networks and multilayered perseptron.