Abstract Background Limited knowledge regarding the relative effectiveness of workplace accident prevention approaches creates barriers to informed decision‐making by policy makers, public health practitioners, workplace, and worker advocates. Objectives The objective of this review was to assess the effectiveness of broad categories of safety interventions in preventing accidents at work. The review aims to compare effects of safety interventions to no intervention, usual activities, or alternative intervention, and if possible, to examine which constituent components of safety intervention programs contribute more strongly to preventing accidents at work in a given setting or context. Date Sources Studies were identified through electronic bibliographic searches, government policy databanks, and Internet search engines. The last search was carried out on July 9, 2015. Gray literature were identified by searching OSH ROM and Google. No language or date restrictions were applied. Searches done between February and July of 2015 included PubMed (1966), Embase (1980), CINAHL (1981), OSH ROM (NIOSHTIC 1977, HSELINE 1977, CIS‐DOC 1974), PsycINFO (1806), EconLit (1969), Web of Science (1969), and ProQuest (1861); dates represent initial availability of each database. Websites of pertinent institutions (NIOSH, Perosh) were also searched. Study Eligibility Criteria, Participants, and Interventions Included studies had to focus on accidents at work, include an evaluation of a safety intervention, and have used injuries at work, or a relevant proxy, as an outcome measure. Experimental, quasi‐experimental, and observational study designs were utilized, including randomized controlled trials (RCTs), controlled before and after (CBA) studies, and observational designs using serial measures (interrupted time series, retrospective cohort designs, and before and after studies using multiple measures). Interventions were classified by approach at the individual or group level, and broad categories based on the prevention approach including modification of: Attitudes (through information and persuasive campaign messaging). Behaviors (through training, incentives, goal setting, feedback/coaching). Physiological condition (by physical training). Climate/norms/culture (by coaching, feedback, modification of safety management/leadership). Structural conditions (including physical environment, engineering, legislation and enforcement, sectorial‐level norms). When combined approaches were used, interventions were termed “multifaceted,” and when an approach(es) is applied to more than one organizational level (e.g., individual, group, and/or organization), it is termed “across levels.” Study Appraisal and Synthesis Methods Narrative report review captured industry (NACE), work setting, participant characteristics, theoretical basis for approach, intervention fidelity, research design, risk of bias, contextual detail, outcomes measures and results. Additional items were extracted for studies with serial measures including approaches to improve internal validity, assessments of reasonable statistical approaches (Effective Practice of Organization of Care [EPOC] criteria) and overall inference. Random‐effects inverse variance weighted meta‐analytic methods were used to synthesize odds ratios, rate ratios, or standardized mean differences for the outcomes for RCT and CBA studies with low or moderate levels of heterogeneity. For studies with greater heterogeneity and those using serial measures, we relied on narrative analyses to synthesize findings. Results In total 100 original studies were included for synthesis analysis, including 16 RCT study designs, 30 CBA study designs, and 54 studies using serial measures (ITS study designs). These studies represented 120 cases of safety interventions. The number of participants included 31,971,908 individuals in 59 safety interventions, 417,693 groups/firms in 35 safety interventions, and 15,505 injuries in 17 safety interventions. Out of the 59 safety interventions, two were evaluating national prevention measures, which alone accounted for 31,667,110 individuals. The remaining nine safety interventions used other types of measures, such as safety exposure, safety observations, gloves or claim rates. Strong evidence supports greater effects being achieved with safety interventions directed toward the group or organization level rather than individual behavior change. Engineering controls are more effective at reducing injuries than other approaches, particularly when engineered changes can be introduced without requiring “decision‐to‐use” by workplaces. Multifaceted approaches combining intervention elements on the organizational level, or across levels, provided moderate to strong effects, in particular when engineering controls were included. Interventions based on firm epidemiologic evidence of causality and a strong conceptual approach were more effective. Effects that are more modest were observed (in short follow‐up) for safety climate interventions, using techniques such as feedback or leadership training to improve safety communication. There was limited evidence for a strong effect at medium‐term with more intense counseling approaches. Evidence supports regulation/legislation as contributing to the prevention of accidents at work, but with lower effect sizes. Enforcement appears to work more consistently, but with smaller effects. In general, the results were consistent with previous systematic reviews of specific types of safety interventions, although the effectiveness of economic incentives to prevent accidents at work was not consistent with our results, and effectiveness of physiological safety intervention was only consistent to some extent. Limitations Acute musculoskeletal injuries and injuries from more long‐time workplace exposures were not always clearly distinguished in research reports. In some studies acute and chronic exposures were mixed, resulting in inevitable misclassification. Of note, the classification of these events also remains problematic in clinical medicine. It was not possible to conduct meta‐analyses on all types of interventions (due to variability in approach, context, and participants). The findings presented for most intervention types are from limited sources, and assessment of publication bias was not possible. These issues are not surprising, given the breadth of the field of occupational safety. To incorporate studies using serial measures, which provide the only source of information for some safety interventions such as legislation, we took a systematic, grounded approach to their review. Rather than requiring more stringent, specific criteria for inclusion of ITS studies, we chose to assess how investigators justified their approach to design and analyses, based on the context in which they were working. We sought to identify measures taken to improve external validity of studies, reasonable statistical inference, as well as an overall appropriate inferential process. We found the process useful and enlightening. Given the new approach, we may have failed to extract points others may find relevant. Similarly, to facilitate the broad nature of this review, we used a novel categorization of safety interventions, which is likely to evolve with additional use. The broad scope of this review and the time and resources available did not allow for contacting authors of original papers or seeking translation of non‐English manuscripts, resulting in a few cases where we did not have sufficient information that may have been possible to obtain from the authors. Conclusions and Implications of Key Findings Our synthesis of the relative effectiveness of workplace safety interventions is in accordance with the Public Health Hierarchy of Hazard Control. Specifically, more effective interventions eliminate risk at the source of the hazard through engineering solutions or the separation of workers from hazards; effects were greater when these control measures worked independently of worker “decision‐to‐use” at the worksite. Interventions based on firm epidemiological evidence of causality and clear theoretical bases for the intervention approach were more effective in preventing injuries. Less effective behavioral approaches were often directed at the prevention of all workplace injuries through a common pathway, such as introducing safety training, without explicitly addressing specific hazards. We caution that this does not mean that training does not play an essential function in worker safety, but rather that it is not effective in the absence of other efforts. Due to the potential to reach large groups of workers through regulation and enforcement, these interventions with relatively modest effects, could have large population‐based effects.