Action observation (AO) is a physical rehabilitation approach that facilitates the occurrence of neural plasticity through the activation of the mirror-neural system, promoting motor recovery in people with stroke. To assess whether AO enhances upper limb motor function in people with stroke. We searched the Cochrane Stroke Group Trials Register (last searched 18 May 2021), the Cochrane Central Register of Controlled Trials (18 May 2021), MEDLINE (1946 to 18 May 2021), Embase (1974 to 18 May 2021), and five additional databases. We also searched trial registries and reference lists. Randomized controlled trials (RCTs) of AO alone or associated with physical practice in adults after stroke. The primary outcome was upper limb (arm and hand) motor function. Secondary outcomes included dependence on activities of daily living (ADL), motor performance, cortical activation, quality of life, and adverse effects. Two review authors independently selected trials according to the predefined inclusion criteria, extracted data, assessed risk of bias using RoB 1, and applied the GRADE approach to assess the certainty of the evidence. The reviews authors contacted trial authors for clarification and missing information. We included 16 trials involving 574 individuals. Most trials provided AO followed by the practice of motor actions. Training varied between 1 day and 8 weeks of therapy, 10 to 90 minutes per session. The time of AO ranged from 1 minute to 10 minutes for each motor action, task or movement observed. The total number of motor actions ranged from 1 to 3. Control comparisons included sham observation, physical therapy, and functional activity practice. AO improved arm function (standardized mean difference (SMD) 0.39, 95% confidence interval (CI) 0.17 to 0.61; 11 trials, 373 participants; low-certainty evidence); and improved hand function (mean difference (MD) 2.76, 95% CI 1.04 to 4.49; 5 trials, 178 participants; low-certainty evidence). AO did not improve ADL performance (SMD 0.37, 95% CI -0.34 to 1.08; 7 trials, 302 participants; very low-certainty evidence), or quality of life (MD 5.52, 95% CI -30.74 to 41.78; 2 trials, 30 participants; very low-certainty evidence). We were unable to pool the other secondary outcomes (motor performance and cortical activation). Only two trials reported adverse events without significant adverse effects. The effects of AO are small for arm function compared to any control group; for hand function the effects are large, but not clinically significant. For both, the certainty of evidence is low. There is no evidence of benefit or detriment from AO on ADL and quality of life of people with stroke; however, the certainty of evidence is very low. As such, our confidence in the effect estimate is limited because it will likely change with future research.