The amount of incoming solar energy that crosses the Earth's atmosphere is called solar radiation. The solar radiation is a series of ultraviolet wavelengths including visible and infrared light. The solar rays at the Earth's surface is one of the key factor in water resources, environmental and agricultural modelling. Solar radiation is rarely measured by weather stations in Iran and other developing countries; as a result, many empirical approaches have been applied to estimate it by using other climatic parameters. In this study, non-linear models, adaptive neuro-fuzzy inference system (ANFIS) and neural network auto-regressive model with exogenous inputs (NN-ARX) along with empirical models, Angstrom and Hargreaves–Samani, have been used to estimate the solar radiation. The data was collected from two synoptic stations with different climatic conditions (Zahedan and Bojnurd) during the period of 5 and 7 years, respectively. These data contain sunshine hours, maximum temperature, minimum temperature, average relative humidity and solar radiation. The Angstrom and Hargreaves–Samani empirical models, respectively, based on sunshine hours and temperature were calibrated and evaluated in both stations. In order to train, test, and validate ANFIS and NNRX models, 60%, 25%, and 15% of the data were applied, respectively. The results of artificial intelligence models were compared with the empirical models. The findings showed that ANFIS (R2=0.90 and 0.97 for Zahedan and Bojnurd, respectively) and NN-ARX (R2=0.89 and 0.96 for Zahedan and Bojnurd, respectively) performed better than the empirical models in estimating daily solar radiation.