Combining multiple classifiers via combining schemes or meta-learners has led to substantial improvements in many classification problems. One of the challenging tasks is to choose appropriate combining schemes and classifiers involved in an ensemble of classifiers. In this paper we propose a novel evidential approach to combining decisions given by multiple classifiers. We develop a novel evidence structure ñ a focal triplet, examine its theoretical properties and establish computational formulations for representing classifier outputs as pieces of evidence to be combined. The evaluations on the effectiveness of the established formalism have been carried out over the data sets of 20-newsgroup and Reuters-21578, demonstrating the advantage of this novel approach in combining classifiers.