The MMSE is the most widely used cognitive test but its accuracy and clinical utility in diagnosing cognitive disorders is not fully known. A meta-analysis of 34 dementia studies and five mild cognitive impairment (MCI) studies was conducted, separated into high and low prevalence settings. In memory clinic settings the MMSE had a pooled sensitivity (Se) of 79.8%, a specificity (Sp) of 81.3%, a positive predictive value (PPV) of 86.3% and a negative predictive value (NPV) of 73.0%. In mixed specialist hospital settings the Se, Sp, PPV and NPV were 71.1%, 95.6%, 94.2% and 76.4%, respectively. In non-clinical community settings the MMSE had a pooled Se of 85.1%, a Sp of 85.5%, a PPV of 34.5% and an NPV of 98.5%. In those studies conducted purely in primary care the Se, Sp, PPV and NPV were 78.4%, 87.8%. 53.6% and 95.7%, respectively. Thus the case-finding ability of the MMSE was best when confirming a suspected diagnosis in specialist settings with correct identification made in 27/30 positive results. It was modestly effective at ruling-out dementia in specialist settings. Conversely, in non-specialist settings, the MMSE was best at ruling out dementia, achieving about 29/30 correct reassurances with less than three false negatives out of every 100 screens. Regarding use of the MMSE in identifying MCI, limited evidence was found with only five robust studies comparing MCI with healthy subjects and three comparing Alzheimer's disease with MCI. Provisionally, the MMSE had very limited value in making a diagnosis of MCI against healthy controls and modest rule-out accuracy. It had similarly limited ability to help identify cases of Alzheimer's disease against MCI. In conclusion the MMSE offers modest accuracy with best value for ruling-out a diagnosis of dementia in community and primary care. For all other used it should be combined with or replaced by other methods.