Use all the possible data to provide prediction all over the year, and also to have situations where Environmental Contingencies occurred. The train and test data sets that were used didn't present this kind of situations. Data can be processed in different ways, for instance, for a 12-hr ahead prediction, the input could be the average of the previous 12 hours, and the output would be the average in the next 12 hours. The same could be applied for all the possible periods of time. Only GRNNs were used to provide the Predictions. The literature mentions the use of Multilayer Perceptrons (MLP) and feed-forward Neural Networks for the Prediction of Pollutant levels. A comparison of different ANNs topologies would be of valuable help. This work is part of a project to implement a Web-based Information Service. This will let people to know the historical data of pollutants, the current situation of pollutants, and to have a forecast of the Pollution.