The purpose of this paper is to define the framework within which empirical investigations of probabilistic grammars can take place and to sketch how this attack can be made. The full presentation of empirical results will be left to other papers. In the detailed empirical work, the author has depended on the collaboration of E. Gammon and A. Moskowitz, and draws on joint work for examples in subsequent sections. Section II presents a simple example of a probabilistic grammar to illustrate the methodology without complications. Section III indicates how such ideas may be applied to the spoken speech of a young child. Because of the difficulties and complexities of working with actual speech, the fourth section illustrates some of the results obtained when the apparatus of analysis is applied to a much simpler corpus, a first-grade reader. The results of an empirical sort in this paper are all preliminary in nature. (Author/AMM)