A study of relevance through comparison of texts judged by users is reported. The relevant set of documents for each information request was characterized by textual properties. The validity of procedures and interpretations is discussed. Consideration is given to possible mechanization of the process and its possible usefulness in development of iterative searching procedures, in building profiles of user needs, and in improving aspects of information retrieval such as question analysis, indexing, thesaurus building, and system evaluation. (Author)