We developed two models for the CO2 separation process by hollow-fiber membrane modules. The explicit model, which is based on mass balances for the separation modules, is compared with the multilayer perceptrons (MLP) back-propagation neural networks model. Experimental data obtained from single-stage module with recycle are used to validate the explicit model as well as to train the MLP neural model. The effectiveness of the model is demonstrated by little discrepancy between experimental data and computational results. The explicit model for the single-stage module can easily be extended to the multi(three)-stage module. Because of the lack of experimental data for multi-stage modules, computational data from the explicit model with and without recycle are used as training data set for the MLP neural model. We examined the effects of recycle on the recovery based on the results of numerical simulations, and could see that the predicting performance is improved by recycle for multi-stage module. From the results of numerical simulations, the proposed models can be effectively used in the analysis and operation of gas separation processes by hollowfiber membrane modules.