During 1919-80, a team of eight teachers and eight researchers at the Center for Research in Writing, working collaboratively, derived a grounded description of the unique and characteristic qualities of writing instruction in the classrooms of the eight teachers. This description was developed through the procedure of progressive coding, which is a method for the continuous analysis of a phenomenon. Progressive coding consists of systematically and repetitively comparing the description of a behavior with the actual behavior and then refining the description to make it conform to the behavior as perceived by the participants. The description of writing instruction in these classes, coupled with an analysis of the institutional context in which the instruction took place, has called into question some common assumptions about writing instruction and the present institutional ways of supporting it. During 1979-80, a team of teachers and researchers at the Center for Research in Writing derived a grounded description of the unique and characteristic qualities of writing instruction in eight elementary school classrooms.1 This description features the teachers' perceptions of the instruction in which they engaged. Framed in an institutional context, it calls into question some of the common assumptions about writing instruction and the present conventional ways of supporting it. Most of the findings of this study resulted from two characteristics of its inquiry, one of which made the other possible. The inquiry was collaborative: the teachers were participants - not subjects, but colleagues of the researchers on the team. And the inquiry was progressive: we engaged in ongoing data analysis, coding information as we gathered it and evaluating our description of behavior by stages. This procedure for evolving a grounded description has particular consequence for writing instruction, since it provides a controlled means of generalizing information beyond the limits of this study; it can be used in framing questions raised by the findings it has already generated and in forming hypotheses about writing instruction for further testing.