OBJECTIVE: The evidence base for social prescribing is inconclusive, and evaluations have been criticised for lacking rigour. This realist review sought to understand how and why social prescribing evaluations work or do not work. Findings from this review will contribute to the development of an evidence-based evaluation framework and reporting standards for social prescribing. DESIGN: A realist review. DATA SOURCES: ASSIA, CINAHL, Embase, Medline, PsycINFO, PubMed, Scopus Online, Social Care Online, Web of Science and grey literature. ELIGIBILITY CRITERIA: Documents reporting on social prescribing evaluations using any methods, published between 1998 and 2020 were included. Documents not reporting findings or lacking detail on methods for data collection and outcomes were excluded. ANALYSIS: Included documents were segregated into subcases based on methodology. Data relating to context, mechanisms and outcomes and the programme theory were extracted and context-mechanism-outcome configurations were developed. Meta-inferences were drawn from all subcases to refine the programme theory. RESULTS: 83 documents contributed to analysis. Generally, studies lacked in-depth descriptions of the methods and evaluation processes employed. A cyclical process of social prescribing evaluation was identified, involving preparation, conducting the study and interpretation. The analysis found that coproduction, alignment, research agency, sequential mixed-methods design and integration of findings all contributed to the development of an acceptable, high-quality social prescribing evaluation design. Context-mechanism-outcome configurations relating to these themes are reported. CONCLUSIONS: To develop the social prescribing evidence base and address gaps in our knowledge about the impact of social prescribing and how it works, evaluations must be high quality and acceptable to stakeholders. Development of an evaluation framework and reporting standards drawing on the findings of this realist review will support this aim. PROSPERO REGISTRATION NUMBER: CRD42020183065.