The objective of this paper is to empirically investigate and compare pricing accuracy of stock index futures models; cost-of-carry model, general equilibriummodel and neural network model. Neural network models could min imize pricingerrors because they do not require unrealistic assumptions, e.g . , underlying assets’stochastic processes. In order to achieve the research objectiv e, we employ threeevaluation statistics; average pricing errors, MAD(Mean Absolut e Deviation) and MSE(Mean Square Error). We find that the general equilibrium model outperformsthe cost-of-carry model and that the neural network model ‘with more lagvariables’performs better than the general equilibrium model. The results imply that not only general equilibrium models but also neural network models could be potential alternatives for more accurate pricing models on stock index futures and that further studies on these models should be warranted.