IMPORTANCE: Childhood maltreatment is associated with mental illness. Researchers, clinicians, and public health professionals use prospective or retrospective measures interchangeably to assess childhood maltreatment, assuming that the 2 measures identify the same individuals. However, this assumption has not been comprehensively tested. OBJECTIVE: To meta-analyze the agreement between prospective and retrospective measures of childhood maltreatment. DATA SOURCES: MEDLINE, PsycINFO, Embase, and Sociological Abstracts were searched for peer-reviewed, English-language articles from inception through January 1, 2018. Search terms included child* maltreatment, child* abuse, child* neglect, child bull*, child* trauma, child* advers*, and early life stress combined with prospective* and cohort. STUDY SELECTION: Studies with prospective measures of childhood maltreatment were first selected. Among the selected studies, those with corresponding retrospective measures of maltreatment were identified. Of 450 studies with prospective measures of childhood maltreatment, 16 had paired retrospective data to compute the Cohen κ coefficient. DATA EXTRACTION AND SYNTHESIS: Multiple investigators independently extracted data according to PRISMA and MOOSE guidelines. Random-effects meta-analyses were used to pool the results and test predictors of heterogeneity. MAIN OUTCOMES AND MEASURES: The primary outcome was the agreement between prospective and retrospective measures of childhood maltreatment, expressed as a κ coefficient. Moderators of agreement were selected a priori and included the measure used for prospective or retrospective assessment of childhood maltreatment, age at retrospective report, sample size, sex distribution, and study quality. RESULTS: Sixteen unique studies including 25 471 unique participants (52.4% female [SD, 10.6%]; mean [SD] age, 30.6 [11.6] years) contained data on the agreement between prospective and retrospective measures of childhood maltreatment. The agreement between prospective and retrospective measures of childhood maltreatment was poor, with κ = 0.19 (95% CI, 0.14-0.24; P < .001). Agreement was higher when retrospective measures of childhood maltreatment were based on interviews rather than questionnaires (Q = 4.1521; df = 1; P = .04) and in studies with smaller samples (Q = 4.2251; df = 1; P = .04). Agreement was not affected by the type of prospective measure used, age at retrospective report, sex distribution of the sample, or study quality. CONCLUSIONS AND RELEVANCE: Prospective and retrospective measures of childhood maltreatment identify different groups of individuals. Therefore, children identified prospectively as having experienced maltreatment may have different risk pathways to mental illness than adults retrospectively reporting childhood maltreatment. Researchers, clinicians, and public health care professionals should recognize these critical measurement differences when conducting research into childhood maltreatment and developing interventions.