The advantages of a hypermedia learning system are the possibility to adapt the content to the learner and to maintain the content easily due to the modular structure. The disadvantages are the well-known problem of the cognitive overhead, and the less discussed problem of little local and nearly no global coherence. The k-med approach is to describe the resources of the system by metadata, connect them by relations and have an ontology containing all relevant concepts. These descriptions enable the system to generate dynamically individual guided tours and table of contents for orientation and navigation according to the preferences of the users. The relations are also used to add small pieces of static text between two resources to connect them.