Objective To evaluate diagnostic accuracy of several relevant cut-off points of Montreal cognitive assessment (MoCA) for mild cognitive impairment (MCI) in Chinese middle-aged adults. Methods Databases including PubMed, EMbase, Web of Science, The Cochrane Library (Issue 5, 2016), OVID, CBM, CNKI, VIP, WanFang Data were searched for diagnostic tests about MoCA for MCI from April 9th 2005 to December 31st 2015. Two reviewers independently screened literatures according to the inclusion and exclusion criteria, extracted data and assessed the methodological quality by QUADAS-2 tool. Then, meta-analysis was performed by Stata 14.0 software. Results A total of 27 studies involving 5 755 participants were included with mean ages from 60 to 80 years old. Among them, 1 997 were diagnosed as MCI patients by Petersen criteria. Based on maximal area under the ROC curve as well as optimal pooled sensitivity and specificity, the optimal cutoff value of MoCA was 25/26, the pooled sensitivity was 0.96 with 95%CI 0.93 to 0.97, specificity was 0.83 with 95%CI 0.75 to 0.89, and DOR was 107 with 95%CI 61 to 188. The subgroup analysis with different research designs, different sources of study participants and different MoCA versions all indicated 25/26 as an optimal cut-off value. Conclusion The optimal cutoff value of MoCA in Chinese middle-aged adults for screening MCI by Petersen criteria was 25/26.