This paper addresses the question of how to qualify a knowledge building dialogue in Web-based collaborative learning. It investigates the idea of qualifying the dialogue through grading requirements and through providing students with meta-awareness around their dialogue. The paper investigates and analyzes the changes in quality of student interaction in three different deliveries of the course that emerge in response to three different types of instruction on expected behavior. From this, preliminary conclusions are drawn on the extent to which descriptions of expected behavior influence the interactive process and the quality of the interaction. The first section provides an introduction. The second and third sections provide some brief information on the Web-based course producing the data of the analysis and an account of the knowledge building perspective and criteria of quality used in the analysis of Web-based dialogues. The fourth section addresses the communicative conditions of Web-based environments, forming part of the rationale behind the hypothesis. The fifth section gives a more detailed account of the research design and of the method used in analysis. The sixth section provides the basic analysis. The seventh section gives a reflection on the results of the analysis, and future research perspectives are discussed. (Contains 24 references.) (AEF)