The complex problem of determining the partition coefficient of the guanidine hydrochloride in aqueous two-phase systems has been less studied. For this reason, an artificial neural network was developed to predict the partition coefficients of guanidine hydrochloride in poly (ethylene glycol) 4000/phosphate/guanidine hydrochloride/water system. The neural model (topology and internal structure) was determined using a neuro-evolutionary technique based on differential evolution algorithm, designed in different variants. This model was able to predict the guanidine hydrochloride concentrations in each phase with a mean relative error of 1.4%, which closely matched the experimental data.