A small-world topology characterizes many complex systems including the structural and functional organization of brain networks. The topology allows simultaneously for local and global efficiency in the interaction of the system constituents. However, it ignores the gradations of interactions commonly quantified by the link weight, w. Here, we identify an integrative weight organization for brain, gene, social, and language networks, in which strong links preferentially occur between nodes with overlapping neighbourhoods and the small-world properties are robust to removal of a large fraction of the weakest links. We also determine local learning rules that dynamically establish such weight organization in response to past activity and capacity demands, while preserving efficient local and global communication.