BACKGROUND: The purpose of this study was to determine 1) the test-retest reliability of adult accelerometer-measured physical activity, and 2) how data processing decisions affect physical activity levels and test-retest reliability. METHODS: 143 people wore the ActiGraph accelerometer for two seven day periods, 1-4 weeks apart. Five algorithms, varying non-wear criteria (20 vs. 60 min of 0 counts) and minimum wear requirements (6 vs. 10 hrs/day for ≥ 4 days) and a separate algorithm requiring ≥ 3 counts per min and ≥ 2 hours per day, were used to process the accelerometer data. RESULTS: Processing the accelerometer data with different algorithms resulted in different levels of counts per day, sedentary, and moderate-to-vigorous physical activity. Reliability correlations were very good to excellent (ICC = 0.70 to 0.90) for almost all algorithms and there were no significant differences between physical activity measures at time one and time two. CONCLUSIONS: This paper presents the first assessment of test-retest reliability of the Actigraph over separate administrations in free-living subjects. The ActiGraph was highly reliable in measuring activity over a seven day period in natural settings but data were sensitive to the algorithms used to process them.