Air pollution is an increasing problem arising from the rapid population growth and economic expansion in Macau over the past decade while serious harmful to human health such as, asthma and bronchitis are caused in parallel. As a result, more and more public awareness has placed on it and an effective system for supervising and forecasting the future Air Pollution Index (API) becomes obviously important. How to build up an accurate and dependable model to predict the future API is the goal of this research. In this study, two different kinds of information can be obtained from Macau Meteorological and Geophysical Bureau (SMG) and they are: historical information (The past daily API records logged from January of 2000 to January of 2008); and meteorological information (Five daily pollutants recorded at the same period such as PM10, SO2, NO2, CO, O3 and five essential weather elements in daily based including temperature, relative humidity, wind speed, solar radiation and pressure). To construct a proper model to describe the API system, we may reasonable think that all the related information should be measured as more information we apply to the model, the better performance it should have. Previous studies show that both Box-Jenkins models and Artificial Neuro-Fuzzy Inference System (ANFIS) models have been widely applied in API forecasting but none of them can be concluded as a universal model in different circumstances because of their common drawback – information singularity. Precisely speaking, through the analyses of historical observations, Box-Jerkins models can use to predict the future API without taking any meteorological information into account. With regard to ANFIS, it is not subject to any historical information, instead, it simply employs the collected meteorological data sets and the actual API values as the input / output pairs and a suitable model can thereby be built for future forecasting after sufficient training. No doubt about it, by applying either model may give inadequate results. Therefore, the hybrid model is developed using the combination of Box-Jerkins model and ANFIS model in order to compensate the shortage of each other. The adopted hybrid model can consider with all the information so that to extend the prediction coverage and improve the forecasting ability. In addition to hybrid approach, we also address the importance of data pre-processing. More specifically, there are over 30,000 observations stored in our historical and meteorological information; missing parts of data seem to be usual. To neglect the missing parts is not recommended since we may sacrifice some information stored behind and consequently, lead to inefficient analyses and bias the results. On the other hand, ten meteorological variables are found in this research and an excessive number of inputs not only impair the transparency of the underlying model, but also increasing the computation complexity. So, try to find out the missing values and figure out the most meaningful parts from all the observations are obviously required. Through different cases analyses, we verify that both missing data handling and input selection are significant and benefitted to the system performance. To demonstrate the utility of the proposed scheme, the hybrid model with data pre-process techniques is used to forecast the daily API values of Macau city in January of 2008. The individual Box-Jerkins model and ANFIS model are also applied in order to assess the performance of the hybrid model. By examining the performance index - root mean square error (RMSE) and mean average percentage error (MAPE), the combined model is proved that it can be an effective way to enhance the forecasting accuracy compared with either the models used separately.