Incident solar radiation is important in agriculture, meteorology, hydrology and climatology studies. However, direct measurements of solar radiation are usually not available where required. Therefore, it is usually estimated by using models based on routine atmospheric variables. In this study, monthly global solar radiation in Mashhad is estimated by multiple linear regression and artificial neural network methods. Atmospheric optical properties, such as cloud fraction, aerosol optical depth, precipitable water and cloud optical depth, derived from MODIS since 2002 to 2015 are used as model input data. According to results, the regression models that use cloud fraction as input variable, are more accurate than others. Comparing neural network with linear regression results for common inputs, shows that error in neural network model is about half of that in linear regression model.