Abstract Background Although supervision is a ubiquitous approach to support health programs and improve health care provider (HCP) performance in low- and middle-income countries (LMICs), quantitative evidence of its effects is unclear. The objectives of this study are to describe the effect of supervision strategies on HCP practices in LMICs and to identify attributes associated with greater effectiveness of routine supervision. Methods We performed a secondary analysis of data on HCP practice outcomes (e.g., percentage of patients correctly treated) from a systematic review on improving HCP performance. The review included controlled trials and interrupted time series studies. We described distributions of effect sizes (defined as percentage-point [%-point] changes) for each supervision strategy. To identify attributes associated with supervision effectiveness, we performed random-effects linear regression modeling and examined studies that directly compared different approaches of routine supervision. Results We analyzed data from 81 studies from 36 countries. For professional HCPs, such as nurses and physicians, primarily working at health facilities, routine supervision (median improvement when compared to controls: 10.7%-points; IQR: 9.9, 27.9) had similar effects on HCP practices as audit with feedback (median improvement: 10.1%-points; IQR: 6.2, 23.7). Two attributes were associated with greater mean effectiveness of routine supervision (p < 0.10): supervisors received supervision (by 8.8–11.5%-points), and supervisors participated in problem-solving with HCPs (by 14.2–20.8%-points). Training for supervisors and use of a checklist during supervision visits were not associated with effectiveness. The effects of supervision frequency (i.e., number of visits per year) and dose (i.e., the number of supervision visits during a study) were unclear. For lay HCPs, the effect of routine supervision was difficult to characterize because few studies existed, and effectiveness in those studies varied considerably. Evidence quality for all findings was low primarily because many studies had a high risk of bias. Conclusions Although evidence is limited, to promote more effective supervision, our study supports supervising supervisors and having supervisors engage in problem-solving with HCPs. Supervision’s integral role in health systems in LMICs justifies a more deliberate research agenda to identify how to deliver supervision to optimize its effect on HCP practices.