We present an access method for set-valued attributes that is based on a multi-level extendible hashing scheme. This scheme avoids exponential directory growth for skewed data and thus generates a much smaller number of subqueries for query sets (up to now fastgrowing directories have prohibited the use of hash-based index structures for set-valued retrieval). We demonstrate the advantages of our scheme over regular extendible hashing both analytically and experimentally. Additionally, we compare the performance of a prototype implementation with that of several other access methods; the results underline the benefits of our approach.