In the present investigation, mathematical models have been developed to optimize hydrogen distribution in the refinery. Five models, Model-0, Model-1, Model-2, Model-3 and Model-4, have been formulated to determine the optimal hydrogen network. Amongst these, Model-0 and Model-1 are NLP networks, whereas the remaining three are MINLP networks. The NLP models are improved gradually to develop MINLP models which incorporate new compressor and PSA. The model considers pressure constraints, source flow balance, sink flow balance, compressor flow balance, sink purity constraint, operating cost, capital cost associated with new equipment, payback period and export cost. Amongst five models, Model-4 is predicted as optimal network which is MINLP model incorporating new compressor and PSA. It predicts reduction in hydrogen by 21.74% and annual profit of $ 16.57 million. The present work selects the optimal type of new compressor based on different capital cost functions. Further, the reliability of the present work is checked through comparison of its results with published models. Copyright (C) 2013, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved.