The objectives of this study were to assess whether Teng's modification of the Mini-Mental State Examination (MMSE) improves its performance as a screening test for cognitive impairment and dementia, and to replicate this comparison in French and English language groups, and for differing assumptions concerning the relative importance of false negative and false positive errors. Screening interviews were conducted with representative samples of people aged 65 or over, set in 36 communities in 10 Canadian provinces. There were 8900 community participants in the Canadian Study of Health and Aging, of whom 1600 also underwent an extensive clinical and neuropsychological examination. Sensitivity, specificity and areas under the receiver operating characteristic (ROC) curve for the original MMSE and modified version (the 3MS) were the main outcome measures. Results are reported for French and English versions of the tests. The results indicate the alpha internal consistency for the 3MS was 0.87, compared to 0.78 for the MMSE. The area under the ROC curve in identifying dementia was 0.93 for the 3MS and 0.89 for the MMSE (p < 0.001). There was less difference between the two tests in identifying all levels of cognitive impairment (AUC 0.80 versus 0.77, p < 0.01). The superiority of the 3MS appears more due to its extended scoring system than to its additional questions. The validity of the MMSE was comparable in English and French samples; results for the 3MS were inconsistent between the two samples, suggesting possible translation problems. In conclusion, the 3MS was superior to the MMSE, justifying the slightly greater burden for its administration and scoring. Neither test worked well in identifying lower levels of cognitive impairment.