The reliability of measures of walking in people with Alzheimer's disease (AD) has not been established despite the increased variability of their walking compared to control groups. The purpose of this study was to examine the test-retest reliability of temporal and spatial gait measurements in community dwelling people with AD. Ten males and 10 females with AD completed 10 walks on an instrumented walkway at self-selected comfortable speed on two occasions, 1 week apart. Data from 10 walks for each subject for each session were averaged and test-retest reliability of velocity, cadence, step length, stride length, step width, toe in/out angle, swing and stance times was measured using intraclass correlation coefficients (ICCs) of the type (3, 1). Coefficients of variation (CV) of method error were calculated to determine between-trial variability and the minimum detectable change (MDC) was calculated to indicate the difference in scores required to show a true change. This analysis was repeated for averages of the first three walks for each subject at each session. ICCs were found to be high (>0.86) for all gait parameters using both 10 and three walks, indicating high test-retest reliability for these measures in people with AD with as few as three walks. Measures of support base and toe in/out angle, however, had higher CV, even with increased numbers of walks, suggesting that only large changes in these parameters can be detected. MDC values suggested that demonstration of real change in gait measures is feasible in this population.