For discriminating amongst various classes, pairwise classifiers are often used in literature. However, the set of optimal variables is selected to be the same for all the pairwise classifiers. This work will investigate the improvement in the classification performance if different sets of optimal variables are selected for different classifiers.