In this paper we argue that the interpretation of discourse involves primarily the selection of semantic models, rather than their construction, as is suggested in a number of recent proposals (e.g., Stenning, 1977; Reichman, 1978). We begin with a modified version of a logical framework developed by van Fraassen (1971) (based on Tarski (1936)) and specifically augment it to allow for relativized quantification (Cushing, 1976) and a dynamic reading of satisfaction rules (Cushing, 1977). We then argue that a selection account of discourse interpretation falls right out of the relativized rule, along with an explanation for at least some "conventional implicatures' (Karttunen and Peters, 1979).