Electrolytic dehumidification based on a polymer electrolyte membrane (PEM) is promising due to the fast humidity control, safe operation and ultra-compactness. However, predictions with existing methods deviated widely, mainly due to the inaccurate calculations of the anode-side over-potential and heat/mass transfer. In this paper, an improved theoretical model was developed, by optimizing the anode-side prediction using newly developed equations for the exchange current density and heat/mass transfer coefficient. The equations were derived by the multi-parameter fitting, with a database obtained by making the simulation results as close as possible to the experimental data under various operating conditions. Compared to the results of previous models, the developed model showed a data trend that was much closer to the experimental data. The overall errors for the moisture removal rate and operating current under most conditions were less than 15% with acceptable average errors of 10.4% and 7.1%, respectively. Particularly, this model showed much higher accuracy at higher air flow rates or larger electrical fields, especially for the current prediction. When the air Reynolds number was above 2 or the applied voltage was above 3 V, the deviation of the prediction could reduce by more than 50% with the new model. And the common problem of overestimation for current prediction (up to 4-5 times) in existing models could be solved. In addition, the parameter analysis of the optimal exchange current density and heat/mass transfer coefficient was conducted. It could be found that the applied voltage during dehumidification had the largest influence. The influences of the anode air inlet temperature, flow rates and inlet relative humidity were also significant. But the effects of cathode air temperature and flow rates were slight. This research significantly improves the prediction accuracy and provides a good guidance for the performance optimization of electrolytic dehumidification systems. (C) 2019 Elsevier Ltd. All rights reserved.