The purpose of this study is to investigate the reliability of artificial neural network (ANN) in air pollution modeling around the industrial port of Sohar, Oman. The emphasis of the study is to predict the concentrations of typical eight air pollutants, namely CO, PM10, NO, NO2, NOx, SO2, H2S, and O3, as a function of the previous day air quality concentrations as well as previous day weather conditions. It also aims to represent the relative contribution of meteorological conditions like wind speed, wind direction, air temperature and relative humidity in the daily variations of ground-level pollutant levels. The ANN models were trained using historical monitoring data of both meteorological and air quality parameters collected during the period of 2006 to 2009. The data were collected from a typical residential area located 700 meters downstream to the industrial port. The ANN models were fed by a total number of 1020 daily basis data sets, divided into a ratio of 3:1 between training and testing sets, respectively. These models were trained based on the multi-layer perceptron (MLP)architecture using back-propagation (BP) algorithm. The ANN models showed very good agreement between actual and predicted values for the different types of pollutants as the coefficient of multiple determinations (R3) was found above 0.70 for all developed models. These results are promising to provide an air pollution prediction system for the city of Sohar. This system makes an early alert, 1-day before, which would alter thousands of residents t air pollution levels that could pose a risk to their health.