Background: Knowledge brokers facilitate the creation, sharing and use of knowledge among different research, policy and practice communities. This paper examines the way collective knowledge brokering can address the syntactic, semantic and pragmatic boundaries that separate research and practice communities. Methods: A comparative ethnographic study of three applied health research projects, each researching the implementation of clinical interventions into healthcare practice. Data was collected through in-depth non-participant observations, semi-structured interviews and documentary analysis with each project team and associated practice stakeholder communities. Findings: The study describes how 'core' project teams and 'peripheral' stakeholder groups can be differentiated as relatively bounded epistemic communities. It also identifies both 'designated' and 'non-designated' knowledge brokers as working within and between these communities. Through comparing the life cycles of these projects, the study identifies the individual and collective brokering activities involved in sharing knowledge among these communities. Discussion: The study reaffirms the idea that it is challenging for a single broker to undertake all brokering tasks, and that knowledge sharing is achieved through sequential and parallel forms of collective knowledge brokering. Furthermore, different brokering strategies are needed to address the syntactic, semantic and pragmatic boundaries that separate epistemic communities. The paper proposes a maturity model of collective knowledge brokering.