Objective: To examine the test—retest reliability of the StepWatch Activity Monitor outputs over two periods, a week apart, in participants with stroke.

Design: Test—retest reliability study over monitoring periods of one, two and three days.

Setting: Participant's usual environment.

Participants: Forty participants more than six months post stroke.

Main measures: StepWatch outputs: total step count, number of steps at high medium and low stepping rates, sustained activity indices, peak activity index.

Results: The intraclass correlation coefficients were high for all StepWatch outputs and all monitoring periods but were highest for the three-day monitoring period (0.930—0.989) and lowest for the one-day monitoring period (0.830—0.950). The coefficient of variation ranged from 6.7% to 48.7% over the monitoring periods, with higher variation shown for shorter monitoring periods. The most reliable four outputs had 95% limits of agreement between three-day periods that were less than 40%. These were total step count (±37.8%), highest step rate in 1 minute (±23.0%), highest step rate in 5 minutes (±38.6%) and peak activity index (±29.8%).

The highest step rate in 1 minute was the only StepWatch output that had 95% limits of agreement less than 40% for the two-day (±31.2%) and one-day (±36.7%) monitoring periods.

Conclusions: Total step count, highest step rate in 1 minute, highest step rate in 5 minutes and peak activity index have good test—retest reliability over a three-day monitoring period, with lower reliability shown by the other StepWatch outputs. In general, monitoring over one or two days is less reliable.