Publish/subscribe systems filter published documents and inform their subscribers about documents matching their interests. Recent systems have focussed on documents or messages sent in XML format. Subscribers have to be familiar with the underlying XML format to create meaningful subscriptions. A service might support several providers with slightly differing formats, e.g., several publishers of books. This makes the definition of a successful subscription almost impossible. This paper proposes the use of an approximative language for subscriptions. We introduce the design of our ApproXFilter algorithm for approximative filtering in a publish/subscribe system. We present the results of our performance analysis of a prototypical implementation.