An approach to cluster characterization is developed. The clusters are constructed by using one of the user-oriented techniques. If the clusters have been characterized on the basis of user preferences, the representations of documents contained within the clusters need not be scanned to determine the relevance of a cluster of documents to a query. Since the cluster representations are decided by the users, this approach results in a higher precision and recall in document retrieval. A program to interview a user automatically to elicit the constructs and the repertory grid has been implemented.