OBJECTIVE The aim of this study was to evaluate the effectiveness and safety of acupuncture for the treatment of post-stroke cognitive impairment (PSCI). METHODS The Cochrane Library, Embase, Medline, China National Knowledge Infrastructure (CNKI), Chinese Science and Technology Periodical (VIP), Wanfang, and Chinese Biological Medicine (CBM) databases were electronically searched from their inception to 10 April 2019. The Montreal Cognitive Assessment (MoCA) scale and Mini-Mental State Examination (MMSE) scale were used as outcomes to assess effectiveness with respect to cognitive function. Assessment of risk of bias (ROB) and Grading of Recommendations Assessment, Development, and Evaluation (GRADE) assessment were performed by two reviewers independently. Data were analyzed using Review Manager (RevMan) 5.3. RESULTS A total of 28 trials with 2144 participants were included in the qualitative synthesis and meta-analysis. Four of the 28 trials (14%) were assessed as being at overall low ROB, 24 of the 28 trials (86%) were assessed as having overall high ROB. The quality of evidence for both MoCA and MMSE were deemed to be very low by the GRADE criteria. Results indicated that acupuncture groups may be benefiting more than non-acupuncture groups with respect to variation of MoCA scores (merged mean difference (MMD): 2.66, 95% confidence interval (CI): 2.18 to 3.13, p < 0.00001; heterogeneity: χ2 = 35.52, p = 0.0007, I2 = 63%), and the heterogeneity decreased in both subgroup analysis and sensitivity analysis. In addition, acupuncture groups might be benefiting more than non-acupuncture groups in terms of changes in MMSE score (MMD = 2.97, 95% CI = 2.13 to 3.80, p < 0.00001; heterogeneity: χ2 = 269.75; p < 0.00001; I2 = 92%), and the heterogeneity decreased in subgroup analysis. Only one RCT addressed adverse events, and the symptoms were mild and did not affect treatment and evaluation. CONCLUSIONS Acupuncture could be effective and safe for PSCI. Nevertheless, the results should be interpreted cautiously due to the high ROB of included trials and very low quality of evidence for assessed outcomes.