In a series of recent studies, we have found that tests from the CogState battery were sensitive to cognitive impairment (i.e., relative to matched controls) in mild to moderate AD, aMCI, and also in healthy older adults who carry putative AD biomarkers (Harel et al., 2011; Lim, Ellis, Harrington, et al., 2012; Lim, Pietrzak, Snyder, Darby, & Maruff, 2012). Of equivalent importance, in a different sample, we also found that performance on these same CogState tasks remained stable despite repeated administration in healthy older adults and was characterized by reliable decline over periods of 12 months or greater in aMCI (Harel et al., 2011). Furthermore, in patients with AD, performance on the verbal list learning task of the CogState battery declined over 1 year (Lim, Pietrzak, et al., 2012). Taken together, these data suggest that the same computerized battery of cognitive tests may be used to measure cognitive function repeatedly in older individuals with normal cognition and in patients with aMCI and AD. However, for use in AD groups, it has been necessary to simplify some (i.e., visual learning and associate learning; Harel et al., 2011) but not all (i.e., verbal learning; Lim, Pietrzak, et al., 2012) tests of memory in order to maximize their acceptability. Finally, the CogState battery has demonstrated sensitivity to cognitive improvement arising from treatment with current pharmacotherapies for AD (e.g., cholinesterase inhibitors), with clinical trial data showing that performance is improved after acute treatment with donepezil in healthy older adults (Pietrzak, Maruff, & Snyder, 2009; Snyder, Bednar, Cromer, & Maruff, 2005) and with daily dosing in AD (Jaeger, Hardemark, & Zettergren, 2011). There is a growing appreciation of the importance of understanding the dynamics and reliability of neuropsychological assessments used for repeated assessments (Duff, 2012; Heilbronner et al., 2010). However, as yet, there has been no direct comparison of the stability and reliability of these same tests between healthy older adults and those with MCI or AD assessed over the same time intervals. The Australian Imaging, Biomarkers, and Lifestyle (AIBL) study is a prospective natural history study of over 1000 adults who are cognitively normal or have either a diagnosis of MCI or mild AD (Ellis et al., 2009). These individuals undergo extensive assessment using psychiatric, neuropsychological, neurological, neuroradiological, and lifestyle measures at 18-month intervals (Ellis et al., 2009; Rowe et al., 2010). The AIBL Rates of Change substudy (hereafter referred to as ROCS) was designed to leverage the care and attention used in recruiting, assessing and characterizing the subjects in AIBL, by taking a subset of each clinical group and assessing them repeatedly at short retest intervals using the CogState battery to determine the extent to which any change in cognitive function could be detected in individuals with different stages of AD over intervals of 1, 2, 3, 6, 12, and 18 months. As the ROCS study is now enrolled fully and complete to the 3-month assessment, these data can be used to examine the acceptability of the tests in healthy adults, and adults with aMCI and AD, as well as to examine the magnitude of differences in performance between these groups. Some clinical trials of putative cognitive enhancers in AD are also conducted over 3 months and these trials generally measure cognitive performance at baseline and then at multiple follow-up assessments (i.e., weeks 4, 8 and 12; Pietrzak et al., 2009; Rogers et al., 1998). Therefore, we also investigated the stability of performance on the battery over 12 weeks in each of these cognitive measures between groups. Data from this prospective study can provide estimates of the expected rate of change in cognitive function over 12 weeks, as well as estimates of associated error (i.e., test–retest reliability and stability of the different outcome measures). Such data can be useful for computing power in clinical trials conducted over the same time interval. Finally, by restricting our analyses to the very short term, the effects of disease-related variability would be minimized, thus allowing direct comparison of any differences in rates of change or stability of performance in the different stages of the disease. The first aim of this study was to directly compare the performance on the CogState battery between healthy older adults, adults with aMCI and adults with AD who had completed 3 months of assessment in the ROCS. The first hypothesis was that for tests where performance could be compared directly, healthy older adults would perform better than adults with aMCI, who would in turn show performance better than adults with AD. The second aim was to determine the stability of performance on the CogState battery over the initial 12 weeks of the ROCS in which it had been administered four times. The second hypothesis was that performance measures on the CogState battery would be reliable and remain stable (i.e., unchanged) over the short test–retest period in healthy, aMCI, and AD groups. The third aim was to explore the estimates of variability in performance over time on the CogState battery and to compare these between the three clinical groups to determine whether they are different.