The model presented in the figure, shows that we input information about the Pollutant that we wish to predict, and Meteorological variables to a black box. This black box contains a Generalized Regression Neural Network (GRNN) that processes the information. This network was trained with information from months of December 2002 and 2003, and it was tested with data from December 2004. The output is the concentration levels of the pollutant in the next, 1, 12 or 24 hours based on the information of the previous hours. To obtain a 1-hr prediction, the Generalized Neural Network (GRNN) needs the information of the hour before. To get a 12-hr prediction, the GRNN needs the information of the previous 12 hours, and so on.