One feature common to Action Research approaches is that they seek to involve the participation of the social actors in the research process. To achieve this, these approaches usually engage, as part of their methodology, in the deconstruction of the image ofscience or at least of that ofthe scientist in order to open up the social field of knowledge production. An immediate question that arises, occasionally in very abrupt ways, for researchers risking themselves along this route, concerns the legitimacy of these new approaches of science working in "real-lire situations". Two areas of legitimacy ofaction research are discussed. First, the construction of "scientific objects" and formulation ofproblems with the partners in the research: these two steps guide the elaboration of the hypotheses for both action and research. Second, the criteria for knowledge validation: we propose that they focus on the quality of the research process rather than on settled results, by considering criteria ofrelevance, fertileness, operationality and learning. The production and validation of scientific knowledge must conform to demanding criteria, even if they are different {rom those applying in the experimental and observation sciences. These criteria derive {rom an evaluation of the quality of the argumentation process as expounded to the social actors.