This study examines automatic indexing performed with a manually constructed thesaurus on a document collection of titles and abstracts of library science master's papers. Errors are identified when the meaning of a posted descriptor, as identified by context in the thesaurus, does not match that of the passage of text which occasioned the posting. Three major types of errors are discussed: term ambiguity, flaws in thesaurus construction, and the automatic word matching procedure. The study concludes with a discussion of reasons for failure of each of the three error types, and an explanation of study limitations. Appendices contain an outline of the hierarchical structure of the thesaurus studied, a random sample of 100 index terms, and other subsets of the thesaurus which support the study. (Author/SW)