This study investigated test–retest functional magnetic resonance imaging (fMRI) reproducibility in 10 healthy older adults and in 10 mild cognitive impairment (MCI) persons using a two-condition (encoding and retrieval) verbal episodic memory task as well as a two-condition (with and without a motor response) phonological processing task. Reproducibility measures included an overlap ratio with four different thresholds, statistical comparisons of the condition contrasts across sessions (test–retest contrasts), ANCOVAs, and intraclass correlation (ICC) on selected regions of interests (ROIs). In all four conditions and for all reproducibility measures, MCI individuals showed fMRI test–retest reproducibility indices that were comparable to those of healthy older adults. At the group level, the comparison of the test–retest condition contrasts yielded very few differences in the areas and level of activation and those differences tended to show a slight reduction of activation in the second session. In addition, the results from the ANCOVAs showed that the fMRI signal measured at the group level does not vary significantly from one session to another. Overlap ratios, however, showed that the fMRI signal failed to produce a reliable pattern of significantly activated voxels across sessions. ICC analyses on selected ROIs indicated that there is high within-subject variability, suggesting reduced reliability at the individual level. Overall, these findings indicate that MCI individuals show fMRI test–retest reproducibility comparable to those of healthy controls and hence that MCI do not alter fMRI reproducibility. Furthermore, they indicate that monitoring treatment effects is reliable when comparing groups but reduced when comparing single individuals. These results have precise implications for the design of longitudinal studies relying on fMRI measures in older adults. Hum Brain Mapp, 2009. © 2009 Wiley-Liss, Inc.