The technical feasibility of a continental information network for astronomy has been demonstrated in the course of a two-month experiment conducted jointly by Dearborn Observatory of Northwestern University and the Stanford University Computation Center. The experiment simulated a scientific information network based on a high-level retrieval language of the non-procedural type named DIRAC. A data-base of astronomical catalogues was maintained in Palo Alto, California, and was queried remotely by a team of astronomers in Illinois. The relevant parameters of approximately one hundred time-sharing sessions were thus recorded. Analysis of the experiments in terms of operating system efficiency, user interface and cost effectiveness supports the idea that the network concept is basic to meaningful scientific documentation systems; it also indicates that generalized software is the key to cost-effective information retrieval in the environment considered and, by extension, in a variety of scientific areas that rely on a combination of bibliographic and catalogued information with a high degree of internal structure. The article reviews the problems of astronomical data structures in their relevance to language design and to the general problem of scientific information handling, and it discusses the various factors: administrative, computational and psychological, that will affect the implementation of future networks. (Author)