Clustering of text documents has been shown to be helpful for representing a high-level picture of large collections [6]. It therefore may have potential to alleviate information overload when people are faced with having to search through large volumes of text data such as World Wide Web, company intranets, or digital libraries. Despite a surge of interest in interactive tools based on clustering on the part of researchers [3] and commercial companies, results of empirical studies of such systems [1][2][6], although as yet few in number, have been negative or inconclusive. This may be at least partially due to the many challenges involved in evaluating interactive retrieval systems [5].