This paper reports on an application of parallel models to the area of information retrieval and argues that massively parallel, distributed models of computation, called connectionist, or parallel distributed processing (PDP) models, offer a new approach to the representation and manipulation of knowledge. Although this document focuses on information retrieval systems used in bibliographic searching, known as document retrieval systems, the discussion generalizes quite readily to broader applications of information retrieval. The retrieval system described makes dynamic use of the internal structure of a database to infer relationships among items in the database. Using these relationships, the system can help overcome incompleteness and imprecision in requests for information, as well as in the database itself. The appendix to this report is a summary of activation rules and parameter values of the test system. (Author/DMC)