A bond-graph model has been developed and applied to analyze the non-steady state transport kinetics of K+ cations in the Donnan dialysis (DD) process. The model is based on a pseudo-thermodynamic network analysis and takes into account all the simultaneously occurring processes and phenomena, i.e. diffusion through aqueous layers, interfacial and membrane ion-exchange kinetics, interdiffusion of K+ and H+ cations, diffusion of a free electrolyte in the membrane, and the osmotic transport of water. Moreover, membrane morphology represented by its heterogeneity and non-uniform distribution of ionogenic groups has been included in the model. The numerical simulations of fluxes demonstrated that the DD efficiency in the investigated system results from osmotic water transport as well as an electrolyte solution sorption into the cation-exchange membrane. These phenomena cause dilution of the stripping phase and enable free diffusion of the electrolyte between the external solutions, respectively. The reverse K+ ions flux from the stripping solution into the feed one observed in DD experiments results from the interdiffusion and non-exchange diffusion of the electrolyte. The numerical calculation results were compared with the experimental data measured for the DD systems operating with a low feed solution concentration (c(f)=0.01 M) and high stripping solution concentration (c(s) up to 1 M) resulting in observable osmotic flux of water. It was found that the developed model can be applied for the prediction of time-dependent concentration in the DD system. Consequently, the prediction of fluxes, concentration factors, recovery factors, etc. is also possible. The model calculations enable also the prediction of the experimental results as dependent on one of the main factors which influence DD efficiency, i.e. the stripping phase concentration.