To test the validity of the StepWatch Activity Monitor (SAM) in subjects with stroke against 2 criterion standards, 3-dimensional gait analysis (3-DGA) and footswitches in a variety of indoor and outdoor walking conditions, including different speeds and different terrains, and to test the accuracy of the SAM when worn on the paretic limb. Criterion standard validation study. Gait laboratory and outside course. Twenty-five participants with physical disability after stroke. Not applicable. The total step count measured simultaneously by SAM and either 3-DGA or footswitches for both paretic and nonparetic limbs. The total step count measured by the SAM and 3-DGA was highly correlated (nonparetic limb, r=.959; paretic limb, r=.896). The 95% limits of agreement (LOA) (derived from Bland-Altman analysis) between the SAM and 3-DGA were within +/-10 steps for SAMs worn on either the nonparetic or paretic limb. The total step count measured simultaneously by the SAM and footswitches was also highly correlated for each limb (nonparetic, r=.999; paretic, r=.963). The 95% LOA between the SAM and footswitches were +/-9 steps on the nonparetic limb but higher at +/-57 steps on the paretic limb. Further analysis showed that the measurement differences occurred during the outdoor component of the combined walk. The 95% LOA between footswitches on both limbs were not more than +/-9 steps for walking, suggesting that the error was accounted for by the SAM on the paretic limb, which both over- and underread the total step count in the outdoor walking conditions. Criterion validity of the SAM to measure steps in both clinical and natural environments has been established when used on the nonparetic limb. However, more errors are apparent when the SAM is worn on the paretic limb while walking over a variety of outdoor terrains. Validation is recommended before use in patients with neurologic conditions affecting bilateral legs because there may be more error, particularly in outdoor activities.