The Clinical Dementia Rating (CDR) Scale comprising global score (CDR-GS) and sum of boxes scores (CDR-SB) is commonly used in staging cognitive impairment; however, its diagnostic accuracy is not well clarified. The meta-analysis aimed to investigate the diagnostic accuracy of the CDR for mild cognitive impairment (MCI) and dementia in older populations. Studies examining the diagnostic accuracy of the CDR for MCI or dementia against reference standards were included from seven electronic databases. The bivariate analysis with a random-effects model was adopted to calculate the pooled sensitivity and specificity of the CDR for MCI and dementia. Fifteen studies investigating the diagnostic accuracy of the CDR-GS (n = 13) or CDR-SB (n = 5) for MCI or dementia were included. The pooled sensitivity and specificity of the CDR-GS for MCI were 93% and 97%, respectively. With respect to dementia, the CDR-GS had superior pooled specificity compared to the CDR-SB (99% vs. 94%), while similar sensitivities were found between the CDR-GS and CDR-SB (both 87%). Significant moderators of an old age, a high educational level, a high prevalence of MCI or dementia, being in a developing country, and a lack of informants' observations may affect the estimation of the sensitivity or specificity of the CDR. Evidence supports the CDR being useful for detecting MCI and dementia; applying the CDR for staging cognitive impairment in at risk populations should be considered. Furthermore, including objective observations from relevant informants or proxies to increase the accuracy of the CDR for dementia is suggested.