Obtaining reliable estimation and information about particulate matters (PM) with aerodynamic diameter less than 10 micron is crucial. This is due to the PM10 serious negative impact on human health and environment. Therefore, studying the concentration and spatial pattern of the PM10 in urban cities, especially populated and industrial cities, has become a popular an important subject for researchers. For this purpose, several pollution stations have been established in different part of cities. These stations have the capability of measuring and recording the concentration amount of various pollutant including PM10, PM2.5, and etc. however, pollution stations can measure the concentration amount of various stations precisely, their measurements are not spatially connected and only spars point observations are provided. One of mostly used and efficient methods to solve this problem is to use remote sensing data. Aerosol optical depth (AOD) and aerosol contribution to apparent reflectance (ACR) are two mostly used remote sensing data which have been used to study the concentration and spatial pattern of PM10. In this study, we have adopted to use ACR images instead of AOD based on three reasons including their higher spatial resolution, spatially connected (without any gap), and at last the absence of aerosol robotic network (AERONET) for AOD retrieved values evaluation in our study area. ACR images can be generated though a relation using red and SWIR (2.1 µm) bands. In particular, we can estimate the surface reflectance (SR) of the Red band from top of atmosphere reflectance (TOAR) of the SWIR (2.1 µm) band. This is possible based on an assumption that the aerosol’s effect on SWIR band is negligible because of its higher wavelength. After the estimation of Red band SR, the difference between Red band SR and TOAR can be an illustration of the amount of atmospheric reflectance. In this study, moderate resolution imaging spectroradiometer (MODIS) level-1B images identified as MOD02HKM with 500 m spatial resolution for 12 days over Tehran, Iran have been utilized. Their corresponding ground measurements of PM10 concentration from 13 pollution stations spread have been used. We have used artificial neural networks to develop the model to estimate the PM10 concentration from ACR data. This has been done in two different approaches including daily modeling (each day separately) and overall modeling (using 12 days data together). Artificial neural networks with one, two, and three hidden layers and sigmoid transfer function using the levenberg-marquardt algorithm are employed to model the relation between ACR values and corresponding PM10 concentration. Also, pollution maps are generated to investigate the spatial pattern of the PM10. In daily modeling artificial neural networks with one, two, and three layer(s) achieved 0.769, 0.806, and 0.848 R² values respectively, representing the higher capabilities of artificial neural networks with more hidden layer. However, it should be noted that the increase in the number of hidden layer will result in overfitting of the model to the training part of data. Finally, the overall modeling using artificial neural networks with one, two, and three hidden layer(s) obtained 0.412, 0.499, 0.503 R² values.