Background: Stroke is a major cause of long-term disability in adults. Several systematic reviews have shown that a higher intensity of training can lead to better functional outcomes after stroke. Currently, the resources in inpatient settings are not always sufficient and innovative methods are necessary to meet these recommendations without increasing healthcare costs. A resource efficient method to augment intensity of training could be to involve caregivers in exercise training. A caregiver-mediated exercise programme has the potential to improve outcomes in terms of body function, activities, and participation in people with stroke. In addition, caregivers are more actively involved in the rehabilitation process, which may increase feelings of empowerment with reduced levels of caregiver burden and could facilitate the transition from rehabilitation facility (in hospital, rehabilitation centre, or nursing home) to home setting. As a consequence, length of stay might be reduced and early supported discharge could be enhanced. Objectives: To determine if caregiver-mediated exercises (CME) improve functional ability and health-related quality of life in people with stroke, and to determine the effect on caregiver burden. Search methods: We searched the Cochrane Stroke Group Trials Register (October 2015), CENTRAL (the Cochrane Library, 2015, Issue 10), MEDLINE (1946 to October 2015), Embase (1980 to December 2015), CINAHL (1982 to December 2015), SPORTDiscus (1985 to December 2015), three additional databases (two in October 2015, one in December 2015), and six additional trial registers (October 2015). We also screened reference lists of relevant publications and contacted authors in the field. Selection criteria: Randomised controlled trials comparing CME to usual care, no intervention, or another intervention as long as it was not caregiver-mediated, aimed at improving motor function in people who have had a stroke. Data collection and analysis: Two review authors independently selected trials. One review author extracted data, and assessed quality and risk of bias, and a second review author cross-checked these data and assessed quality. We determined the quality of the evidence using GRADE. The small number of included studies limited the pre-planned analyses. Main results: We included nine trials about CME, of which six trials with 333 patient-caregiver couples were included in the meta-analysis. The small number of studies, participants, and a variety of outcome measures rendered summarising and combining of data in meta-analysis difficult. In addition, in some studies, CME was the only intervention (CME-core), whereas in other studies, caregivers provided another, existing intervention, such as constraint-induced movement therapy. For trials in the latter category, it was difficult to separate the effects of CME from the effects of the other intervention. We found no significant effect of CME on basic ADL when pooling all trial data post intervention (4 studies; standardised mean difference (SMD) 0.21, 95% confidence interval (CI) -0.02 to 0.44; P = 0.07; moderate-quality evidence) or at follow-up (2 studies; mean difference (MD) 2.69, 95% CI -8.18 to 13.55; P = 0.63; low-quality evidence). In addition, we found no significant effects of CME on extended ADL at post intervention (two studies; SMD 0.07, 95% CI -0.21 to 0.35; P = 0.64; low-quality evidence) or at follow-up (2 studies; SMD 0.11, 95% CI -0.17 to 0.39; P = 0.45; low-quality evidence). Caregiver burden did not increase at the end of the intervention (2 studies; SMD -0.04, 95% CI -0.45 to 0.37; P = 0.86; moderate-quality evidence) or at follow-up (1 study; MD 0.60, 95% CI -0.71 to 1.91; P = 0.37; very low-quality evidence). At the end of intervention, CME significantly improved the secondary outcomes of standing balance (3 studies; SMD 0.53, 95% CI 0.19 to 0.87; P = 0.002; low-quality evidence) and quality of life (1 study; physical functioning: MD 12.40, 95% CI 1.67 to 23.13; P = 0.02; mobility: MD 18.20, 95% CI 7.54 to 28.86; P = 0.0008; general recovery: MD 15.10, 95% CI 8.44 to 21.76; P < 0.00001; very low-quality evidence). At follow-up, we found a significant effect in favour of CME for Six-Minute Walking Test distance (1 study; MD 109.50 m, 95% CI 17.12 to 201.88; P = 0.02; very low-quality evidence). We also found a significant effect in favour of the control group at the end of intervention, regarding performance time on the Wolf Motor Function test (2 studies; MD -1.72, 95% CI -2.23 to -1.21; P < 0.00001; low-quality evidence). We found no significant effects for the other secondary outcomes (i.e. patient: motor impairment, upper limb function, mood, fatigue, length of stay and adverse events; caregiver: mood and quality of life). In contrast to the primary analysis, sensitivity analysis of CME-core showed a significant effect of CME on basic ADL post intervention (2 studies; MD 9.45, 95% CI 2.11 to 16.78; P = 0.01; moderate-quality evidence). The methodological quality of the included trials and variability in interventions (e.g. content, timing, and duration), affected the validity and generalisability of these observed results. Authors' conclusions: There is very low- to moderate-quality evidence that CME may be a valuable intervention to augment the pallet of therapeutic options for stroke rehabilitation. Included studies were small, heterogeneous, and some trials had an unclear or high risk of bias. Future high-quality research should determine whether CME interventions are (cost-)effective.