In the present study, the Kalman filter algorithm was applied to forecast the wintertime PM10 concentrations of Macau. The algorithm was implemented on an AR(2) model and an AREX model, respectively. The AR(2) model is essentially an autoregressive model of order 2, i.e., the daily averaged PM10 concentration tomorrow is predicted by a linear combination of the PM10 concentrations in the previous two days. The AREX model is built based on the AR(2) model. It is a combination of the autoregressive model and the exogenous inputs such as the wind speed and the wind direction on the day of prediction. Both models were tested by using the PM10 concentrations and the meteorological data between November of 2004 and February of 2005. It was found that the mean absolute prediction error percentage of the AR(2) model was 36.36%, with an RMS error of 34.94 µg m-3. The Pearson correlation coefficient between the predictions and the measurements is 0.59.Time-delay problem was associated with the AR(2) model, i.e., the trend of the predicted PM10 concentrations generally lagged behind the trend of the measurements. On the other hand, the error percentage of the AREX model was 32.45%, with an RMS error of 27.08 µg m-3. The Pearson correlation coefficient is 0.75. The time-delay problem was improved and the trend of the predictions was in good agreement with the measurements. The AREX model outperformed the AR(2) model since the meteorological conditions could reflect the dispersion condition and the nature of the replenishing air masses on the day of prediction. It was concluded that the Kalman filter was promising in the air quality prediction but caution should be made in the selection of the model classes.