The augmented transition network described in this report was developed in the course of work in semantic interpretation in the context of a computer system which answers English questions. In order to provide mechanical input for the semantic interpreter, a parsing program based on the notion of a "recursive transition network grammar" was developed. The form of presentation of rules made possible by this grammar is called a "recursive transition network," the augmented version of which is presented here. The parsing system has proved to be an extremely powerful system capable of performing the equivalent of transformational analysis in little more time than that customarily required for context free analysis alone. The system is also convenient for the designer of the grammar and facilitates experiments with various types of structural representations and various parsing strategies. This report, the first of several, presents a discussion of the augmented transition network as a grammar model, including a number of theoretical results concerning the efficiency of the model for parsing. (FWB)