BACKGROUND AND PURPOSE. Evaluation of physical therapy for gait disorders in patients with Parkinson's disease (PD) requires an under-standing of how the patients' medication cycle affects function. Four experiments were conducted to investigate stability of gait variables. METHODS AND RESULTS. In experiment 1, 15 subjects with idiopathic PD and 15 subjects without PD performed two sets of gait trials spaced 30 minutes apart, with the initial trial conducted with the subjects at a peak dose of medication. Strong correlations, as determined by intraclass correlation coefficients (ICC[2,1]), occurred between repeat measures of speed (ICC = .92), cadence (ICC = .92), stride length (ICC = .94), and time spent in double support (DS) (ICC = .93). In experiment 2, 16 subjects with PD were tested at the same time on two consecutive days. There was good repeatability for speed (ICC = .88), cadence (ICC = .85), stride length (ICC = .84), and DS (ICC = .80). When we assessed the repeatability of measurements taken at peak dose and at end of dose, there were low correlations for speed (ICC = -.54), cadence (ICC = -.07), stride length (ICC = -.35), and DS (ICC = -.38). In a final experiment on 16 subjects with PD, we used time-series analysis to examine the stability of measurements taken every 15 minutes for 2 1/2 hours. Slopes of regression models, standard deviations, and residual autocorrelations were negligible, indicating that the measurements were stable. CONCLUSION AND DISCUSSION. The parkinsonian gait pattern is reproducible across either brief time intervals or 24 hours when peak medication prevails. At the end of dose, however, marked changes in gait occur, apparently related to depletion of medication.