Genetic alterations driving cancer are known to be spread over a large number of genes. Deciphering driver alterations from passenger alterations that may however be selected by single gene analysis is a major challenge. Alterations that characterize a given subtype of cancer are of particular interest. However, characterizing alteration sets rather then handling single alterations is a difficult task because of the combinatorial explosion of the number of sets. We consider a set of gene amplifications, deletions or mutations in tumor samples for which the subtypes of a given cancer are known. We consider that an alteration set characterizes a given subtype with respect to the others if they are frequent in that given subtype and rare for the others. We propose an efficient algorithm that outputs a ranked list of such alteration sets or pathways. The relevance of the output is illustrated using alteration data on bladder cancer.