We examined the degree of interrater agreement on the Modified Mini-Mental State Examination (3MS), administered both in the home and at the clinical examination, to determine the boundaries of reliable individual changes for 257 community-dwelling older persons who received a diagnosis of dementia at CSHA-1. Individual score differences were approximately normally distributed (mean of differences 0.2; SD 8.0; 95% confidence interval -16 to 16). The intraclass correlation coefficient was 0.85. Except for the language of testing, there was no relationship between score differences and the determinants investigated (i.e., age, education, type and severity of dementia). This study provides evidence that, in a time frame compatible with no change in cognition, the discrepancy between repeat 3MS scores can be as large as ± 16. These limits represent the range of variability consistent with no change and should be considered when interpreting individual change scores.