Summary Musculoskeletal disorders (MSDs) are one of the main causes of work disability. Several types of work rehabilitation programs have proven effective in recent decades, yet their level of implementation remains low. One possible explanation is that these programs are complex and have to account for highly varied social, cultural, and legal contexts. Thus it is not always clear which components of the programs are effective (black box phenomenon) or what the underlying theories are. There is a real need to better understand what works in these programs, for whom, in what contexts, why, and how. This study therefore had three specific objectives: (1) to explain which contextual aspects and which program components work, for whom, and under what circumstances; (2) to define the theories underlying work rehabilitation programs; and (3) to put forward pragmatic recommendations for use by the various stakeholders involved in these programs. A realist review was conducted of a number of work rehabilitation programs that included a workplace intervention. Using this method, three main elements were found to characterize the programs: context (C), which refers to the characteristics of the program implementation conditions and the program components; mechanisms (M), which describe what allows a program to achieve its outcomes; and outcomes (O), which are the effects – both expected and unexpected – of a program, resulting from the interaction among different mechanisms in different contexts. The original studies retained for this research (n=8) consisted of the controlled trials included in a systematic review (van Oostrom et al., 2009) and its recent update (van Vilsteren et al., 2015). Articles providing complementary information on these studies completed the corpus (n=16). A total of 24 articles were analyzed in depth. Some 50 configurations were developed in an effort to pinpoint the probable associations between contexts, mechanisms, and outcomes. Atlas.ti qualitative analysis software was used. Based on the CMO configurations developed, a search for demi-regularities brought points of convergence to light. Lastly, a Web-based survey on the clarity, pertinence, usefulness, exhaustiveness, and feasibility of the recommendations generated by the study results was administered to 31 potential users. Regarding Objective 1, five macro components of interventions as well as the specific components of each macro component were identified for various types of workers (manual or sedentary), regardless of the site (back or upper extremity) or the phase of their MSD (acute, sub-acute, or chronic). These five components were: (1) assessment (e.g. overall assessment, taking into account the various dimensions of the individual and including questions about the work situation), (2) timeframe (workplace intervention beginning quickly, following the start of rehabilitation), (3) stakeholder diversity (e.g. active involvement of stakeholders from different systems, participation of an actor from outside the organization to try to bring the varying interests together), (4) information sharing (e.g. intersectoral coordination by a person responsible for communications and actions pertaining to management of worker records, general agreement among the stakeholders involved), and (5) accommodations (e.g. adoption of a problem-solving approach, direct supervisor’s participation in solution implementation). However, certain contextual aspects appear crucial if these components are to produce positive outcomes: the work activity analysis skills needed by a professional close to or integrated into the clinical team, incentives to induce organizations to collaborate in an early workplace-based intervention, and an intersectoral coordinator who actively cultivates and maintains close, frequent, and effective relationships with all parties concerned. Regarding Objective 2, six main groups of mechanisms encompassing psychoaffective, cognitive, and identity-related dimensions were identified. These were: (1) reassurance of the worker, (2) worker satisfaction, (3) preservation of the worker’s identity, (4) coherence between the worker’s needs and the interventions, (5) the proactive behaviour and active involvement of the stakeholders, and (6) the workplace stakeholders’ perception of the pressure exerted on them. The proximal effects, namely, the mechanisms closest to the return-to-work objective, were also divided into eight large groups based on their affinities; they too concerned similar dimensions. It is important to understand that all these groups of mechanisms and proximal effects were transversal, i.e. they were deployed in more than one macro component at a time. Regarding Objective 3, the participants considered most of the proposed recommendations to be clear, pertinent, useful, and exhaustive. However, they regarded all the recommendations, except for those associated with the macro component of assessment, as difficult to implement in practice settings. Taken separately, the results of this study are corroborated by numerous scientific articles. However, our study is original in its assertion that what works necessarily requires intersectoral action among professionals and practitioners from different sectors (healthcare, workplace, insurance). This action must be closely coordinated and requires leadership that is shared by stakeholders in order to achieve the return-to-work objective. From this perspective, additional research is needed to identify more clearly the conditions allowing optimal deployment of such intersectoral action. Our results further reveal an interweaving of the mechanisms and are associated more closely with interventions carried out in a workplace than a clinical setting. These mechanisms therefore concur perfectly with the perspective that regards workers’ experimentation with a variety of situations in their workplace as crucial and as having a direct influence on their representations, and ultimately, on their actions. These results are consistent with the self-regulation model of Leventhal et al. (1980, 2013) and the work of Coutu et al. (2007, 2010, 2011), while veering away from the strictly medical model of MSD case management. Even so, it remains impossible to assert that, to produce the desired outcome, all components of the intervention must be present simultaneously in certain contexts or that all the mechanisms must be in play concurrently. Nor can the return to work be seen simply as the product of a linear process. It must instead be seen as the result of dynamic processes generating feedback loops and numerous interactions. This ripple effect concept illustrates the interdependence and synergy between the intervention components and mechanisms in certain contexts. Three possible avenues emerge for future research: (1) enhancement of the intervention components that work, along with their specific contextual aspects, by continuing to use the same method but broadening the inclusion criteria to more than one type of study design; (2) further investigation of the mechanisms at play, based on work done in psychology, sociology, and anthropology; (3) experimentation with and assessment of intervention components that work, while taking contextual aspects into account, among a different worker population such as that with common mental disorders. In conclusion - and despite the seemingly major challenges it poses - it is essential to recognize the interdependence of the various sectors involved in the return to work and their reciprocal influences, and if necessary, to orchestrate their involvement. This finding confirms the eminently complex, dynamic, and social nature of all interventions aimed at promoting the return to work of workers with an MSD.