As metadata is often not sufficiently provided by authors of Learning Resources, automatic metadata generation methods are used to create metadata afterwards. One kind of metadata is categorization, particularly the partition of Learning Resources into distinct subject categories. A disadvantage of state-of-the-art categorization methods is that they require corpora of sample Learning Resources. Unfortunately, large corpora of well-labeled Learning Resources are rare. This paper presents a new approach for the task of subject categorization of Learning Resources. Instead of using typical Learning Resources, the free encyclopedia Wikipedia is applied as training corpus. The approach presented in this paper is to apply the k-Nearest-Neighbors method for comparing a Learning Resource to Wikipedia articles. Different parameters have been evaluated regarding their impact on the categorization performance.