An attempt is made to present a robust and reliable empirical correlation based on a wide range of data sets to predict gas viscosity using pressure, temperature, and density of gas. Using an accurate value for gas viscosity at any range of operational pressure and temperature is important to simulate gas flow behavior properly. In this study, a novel method is developed using artificial neural network, statistical techniques, and nonlinear optimization to predict hydrocarbon gas viscosity. First, a data set from performed pressure-volume-temperature (PVT) and chromatography tests on Iranian gas reservoirs are gathered and added to prepaid data sets from literature to maximize the validity range of the correlation. Then, the important factors are selected using an artificial neural network. Afterward, the correlation was developed using multivariable regression and nonlinear optimization. Furthermore, the validation of this correlation was approved by drawing predicted gas viscosity versus pressure. Results also proved that the obtained correlation has more accuracy compared to other ones for a randomly selected test data set.