This paper deals with the problem of modelling air pollution time series recorded at a given point by using a multi-layer perceptron based approach. In particular it is shown that the traditional back propagation (BP) algorithm, widely considered in the literature for this kind of application, can be improved to predict episodes of poor air quality. A novel version of the BP algorithm is developed and its performance is compared with that of the traditional algorithm in a case study which refers to the modelling of 1 h average maximum daily ozone concentration recorded in the industrial area of Melilli (Siracusa, Italy). The results obtained so far indicate that the proposed algorithm outperforms the traditional BP algorithm.