Protein domain structure classification systems such as CATH and SCOP provide a useful way to describe evolutionary structure-function relationships. Similarly, the Pfam sequence-based classification identifies sequence-function relationships. Nonetheless, there is no complete direct mapping from one classification to another. This means that functional annotations that have been assigned to one classification cannot always be assigned to another. Here, we present a novel content-based filtering approach called CAPS (Computing direct Associations between annotations of Protein Sequences and Structures) to systematically analyze multiple protein-domain relationships in the SIFTS and UniProt databases in order to infer direct mappings between CATH superfamilies, Pfam clans or families, and SCOP superfamilies. We then compare the result with existing mappings in Pfam, InterPro, and Genome3D.