We have presented a novel approach for the integration of distributed self-organizing maps into higher-order maps by using the weight vectors of trained SOMs as input data to create a referencing map. Combining several independently managed maps offers itself to cope with situations where a number of distributed sources of information need to be managed independently, like it is the case in the digital library arena. Multiple small maps offer the additional advantage of being easier to retrain, thus providing greater flexibility, which is especially important in dynamically changing environments. Integrating several small SOMs allows the creation of large systems while keeping the overhead and maintenance cost low due to the distributed structure. The referencing SOMs need not necessarily form a strict hierarchy. Rather, selected maps of various orders can be integrated into any single map, allowing users to build their personal library systems including maps or even portions of maps that are relevant to their needs. Since the principle of combining these distributed sources are based on a rather generic vector-oriented representation, different types of systems may be integrated, providing a high degree of flexibility.