The volumetric efficiency of gaseous mixture (VEGM) is regarded as a function of speed and intake manifold absolute pressure of a pre-mixed spark ignition coal-bed gas engine. Three models of VEGM were developed based on polynomial, BP neural network and the adaptive neural fuzzy inference system (ANFIS), respectively, by using the system identification method and the experimental data. The modeling efficiencies of various models was compared. In order to validate the models, the volumetric efficiency models were embedded into the mean value model of the coal-bed gas engine respectively. The estimated output of the mean value model was compared with the experiments data. The results show that non-parametric models of the volumetric efficiency are more accurate than the parametric model for prediction and more suitable for system simulation as a sub-model.