A major shortcoming of polynomial approximation in the modelling of distillation columns is the difficulty encountered while choosing the number and location of collocation points, which are usually done by rule of the thumb, inevitably giving rise to high dimensionality and longer computation time for the resulting model. In order to take full advantage of polynomial approximation in the modelling of complicated multicomponent distillation columns, modifications must be made to the model reduction procedure originally proposed by Cho. This is achieved by putting in special polynomials to each of the variable profiles. Furthermore, the number and location of the collocation points can be determined by the optimization of an appropriate objective function. This would bring about less dimensionality and less computation time for the resulting reduced-order model as compared with Cho＇s procedure while its accuracy is still kept excellent. The effectiveness of such modifications is illustrated by two simulation examples. Both order reduction and accuracy improvement are remarkably indicated for steady state and dynamic state simulations of distillation columns.