Drug repositioning, an innovative therapeutic application of an old drug, has received much attention as a particularly costeffective strategy in drug R&D. Recent work has indicated that repositioning can be promoted by utilizing a wide range of information sources, including medicinal chemical, target, mechanism, main and side-effect-related information, and also bibliometric and taxonomical fingerprints, signatures and knowledge bases. This article describes the adaptation of a conceptually novel, more efficient approach for the identification of new possible therapeutic applications of approved drugs and drug candidates, based on a kernel-based data fusion method. This strategy includes (1) the potentially multiple representation of information sources, (2) the automated weighting and statistically optimal combination of information sources, and (3) the automated weighting of parts of the query compounds. The performance was systematically evaluated by using Anatomical Therapeutic Chemical Classification System classes in a cross-validation framework. The results confirmed that kernel-based data fusion can integrate heterogeneous information sources significantly better than standard rank-based fusion can, and this method provides a unique solution for repositioning; it can also be utilized for de novo drug discovery. The advantages of kernel-based data fusion are illustrated with examples and open problems that are particularly relevant for pharmaceutical applications.