A generalized compartmental modelling method for the derivation of reduced-order models for multicomponent staged separation columns is presented. In contrast to the one proposed by Benallou, this method uses dynamic state relationships, rather than steady state ones, such that superfluous composition variables resulting from the compartmental representation of staged separation columns may be eliminated. The accuracy of transient responses of the resulting model is, therefore, substantially improved. Furthermore, the phenomena of initial inverse responses, usually encountered in the development of reduced-order methods, has also been deleted completely. Simulations of a deethanizer have been conducted to evaluate this modelling method. Results indicate high efficiency and robustness in minimizing the dimensionality and computation time of the staged separation column model.