Background. Upper limb (UL) impairment is among the most persistent and functionally consequential sequelae of stroke, and a large proportion of survivors do not regain useful arm and hand function. Task-oriented training (TOT) — the repeated, goal-directed practice of functionally meaningful activities, structured according to motor learning principles — has become the dominant paradigm in contemporary UL neurorehabilitation. Its magnitude of benefit, and the conditions under which that benefit is realised, remain contested.

Objective. To systematically identify, appraise and quantitatively synthesise the best available evidence on the effect of TOT on UL motor impairment, arm and hand function, and activities of daily living (ADL) after stroke, and to examine dose, training type and timing as moderators of effect.

Methods. A systematic search of MEDLINE/PubMed, the Cochrane Library (CDSR and CENTRAL), Embase, CINAHL, PEDro and Google Scholar was conducted for randomised controlled trials (RCTs) and high-quality systematic reviews with meta-analysis evaluating repetitive, task-specific or task-oriented UL training in adults after stroke. Reporting follows the PRISMA 2020 statement. Because valid de novo pooling requires participant-level summary data that were not uniformly retrievable, the quantitative component of this review synthesises published pooled effect estimates from methodologically robust meta-analyses, together with narrative appraisal of landmark RCTs. This design and its consequences are stated explicitly rather than concealed.

Results. The consolidated evidence base comprises 33 RCTs with 36 intervention–control pairs and 1,853 participants in the definitive Cochrane review of repetitive task training, together with 467 RCTs (N = 25,373) in the largest umbrella synthesis of physical therapy after stroke. Pooled estimates favour TOT for arm function (standardised mean difference [SMD] 0.25; 95% CI 0.01–0.49; 11 studies; n = 749), hand function (SMD 0.25; 95% CI 0.00–0.51; 8 studies; n = 619) and ADL (SMD 0.28; 95% CI 0.10–0.45). Effects were moderated by training type: mixed, multi-task programmes yielded SMD 0.42 (95% CI 0.17–0.67; 17 trials), whereas single-task (SMD 0.07; 95% CI −0.42–0.55) and whole-therapy approaches (SMD 0.10; 95% CI −0.24–0.43) did not reach significance. Higher practice dose conferred additional benefit (g = 0.35; 95% CI 0.26–0.45). GRADE certainty was low to moderate, limited principally by unclear risk of bias from incomplete reporting.

Conclusions. TOT produces small but consistent and clinically defensible improvements in UL function and ADL after stroke, and is superior to neurofacilitation-based comparators. The benefit is conditional: it depends on varied, adequately dosed, genuinely task-specific practice rather than on the label 'task-oriented'. Dose-matched trials such as ICARE indicate that when usual care already delivers structured functional practice, further structuring adds little. Future trials should report dose in repetitions rather than scheduled minutes and stratify by baseline severity and corticospinal tract integrity.