Parallel connection is usually adapted in fuel cell system for the high power applications. Fuel cell subsystem has different dynamic characteristics. The connection will show imbalance in load sharing and current distribution. Thus a robust parallel operation and current sharing algorithm is required. In this paper the block diagram of parallel fuel cell generation system is presented, and a master-slave model predictive control (MPC) method is proposed for the current-sharing control of the systems. Neural network optimization is used to reduce the computational time of single step optimization in model predictive control. The proposed MPC-based current sharing strategy is verified to be feasible and effective. Copyright (C) 2015, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved.