One of the major airborne pollutants in urban environments is Particulate Matter (PM) containing inhalable particles that penetrate the thoracic region of the respiratory system determining considerable negative health e ects, which aggravates with the lower sizes of particles, exposure duration and people Rs vulnerability(age, medical record, socio-economical status). We developed an on-line monitoring system for PM2:5 ( ne particulates) which uses self-designed microstations with an integrated early warning mechanism based on a neural network with a wavelet decomposition preprocessing. Using the Daubechies db3 wavelets as a decomposing preprocessor of hourly averages time series of PM2:5 has signi cantly improved the out of sample forecasted values compared to the soled use of FANN.