The significance of automatic document summarization increases with the threat of information overload we are facing. Short summaries can be presented to users, for example, in place of fulllength documents found by a search engine in response to a user’s query. We have analyzed variousapproaches to document summarization, using some existing algorithms and combining these with a novel use of itemsets. The resulting summarizer is evaluated by comparing classification of original documents and that of abstracts generated automatically. Despite highly promising results achieved by this evaluation, readability of abstracts must be further improved by integrating additional heuristic approaches.