This paper presents a comparative study of four schemes for the soft fusion of the outputs of multiple classifiers. In all the approaches, the weights assigned to the classifiers or groups of them are determined by a quadratic programming technique. The approaches used are: 1) weighted averaging; 2) combination via the Choquet integral with the fuzzy measures; 3) linear combination of order statistics; and 4) combination via the Choquet integral with full-fuzzy measure. An empirical evaluation using widely accessible data sets substantiates the usefulness of the aggregation schemes based on the fuzzy integral.