The act of referencing another author's work in a scholarly or research paper is usually assumed to signal a direct semantic relationship between the citing and cited work. The present article reports a study that examines this assumption directly. The purpose of the research is to investigate the semantic relationship between citing and cited documents for a sample of document pairs in three journals in library and information science: Library Journal, College and Research Libraries, and Journal of the American Society for Information Science. A macroanalysis, based on a comparison of the Library of Congress class numbers assigned citing and cited documents, and a microanalysis, based on a comparison of descriptors assigned citing and cited documents by three indexing and abstracting journals, ERIC, LISA, and Library Literature, were conducted. Both analyses suggest that the subject similarity among pairs of cited and citing documents is typically very small, supporting a subjective, psychological view of relevance and a trial‐and‐error, heuristic understanding of the information search and research processes. The results of the study have implications for collection development, for an understanding of psychological relevance, and for the results of doing information retrieval using cited references. Several intriguing methodological questions are raised for future research, including the role of indexing depth, specificity, and quality on the measurement of document similarity. © 1993 John Wiley & Sons, Inc.