We present an approach for the disambiguation of textual mentions of ambiguous names: disambiguation means here the identification of the true entity denoted by a name phrase appearing in a query context through its assignment to the corresponding Wikipedia article. If this article does not exist, we assign this query to a default entity. Ambiguity of names is a major problem in information retrieval and causes uncertainty in the assignment of name phrases to existing knowledge base entries. We propose a kernel classifier to approach this problem and compare two Wikipedia structures to construct a rich feature space. The first approach relies on Wikipedia categories, the second on relations constructed from Wikipedia's hyper link structure. We evaluate both approaches on the German version of Wikipedia and show that both outperform a baseline approach using simple cosine similarity.