Gasoline blending is an important unit operation in refineries. A reliable model for the blending process is beneficial for the plant operation and prediction of gasoline qualities. Since the blending process does not follow the ideal mixing rule in practice, Artificial Neural Network (ANN) models have been developed to determine the Research Octane Number (RON) of the gasoline blends produced in Tabriz refinery. The developed ANN models employ the volumetric content of six most commonly used fractions in gasoline productions multiplied by their octane number as input variables. Different ANN models were examined and the optimum model was selected to be compared with a multiple regression model available in the literature. Genetic Algorithm (GA) technique was utilized to estimate the regression model parameters. Results show that the ANN model simulated the gasoline blending process much better than the regression model as judged by the higher value of R² 0.9812 vs. 0.9538), lower value of MSE (0.0094 vs. 0.0238), and lower value of AARE (0.910 vs. 0.1464).