A study was launched in 2005 to systematize and compare a series of ten contrasting past and on-going experiences in which research has been conducted with local actors (such as farmers and farmers' organizations, but also extension services, governments, private sector, etc.). The main objectives of this study are (i) to draw lessons in terms of research approaches, modalities, methods, tools, and results and (ii) to propose guidelines to improve the design and conduct of research projects focusing on the conception of innovations in partnership among multiple stakeholders. Cross-analysis of the case studies was pursued in three directions: (1) the balance reached between problem resolution, knowledge generation and empowerment of local actors, (2) the formalization of partnerships and (3) the modalities adopted for steering activities and for partnership governance. Assessing the distance between actual project practices followed in each experience, and what could be called an "ideal" action-research process, rigorously based on established A-R concepts, lies at the heart of this analysis. Preliminary results confirm that each experience is the result of an encounter among specific individuals, who purposefully broke away from paradigms reigning locally or institutionally for effecting change. It also shows that the research and innovation process is not a linear process with different well planned phases and cycles, but results from how projects deal with tensions between stakeholders and how they generate the adjustments necessary to achieve eventual success in problem-solving and generate knowledge. The diverse set-ups co-constructed among stakeholders are not only means to achieve common objectives, but they also embody high-stake challenges throughout the project life. To keep the spirit of the partnership, it may be useful to formalize mutual commitments: this helps to avoid the disruptive consequences of the corresponding power struggles, and eventually del