Abstract: This study was designed to test the hypothesis that there is no significant difference in selected temporal and linear stride variables of the extended canter in horses bred and trained for dressage or racing. Nine advanced-level dressage horses and 7 Thoroughbred racehorses were filmed at a frame rate of 200 Hz at an extended canter on a sand track. Two strides were recorded per trial, and each horse performed 6 or 7 trials. Temporal and linear data were determined from the films, and descriptive statistics (mean, SD) were calculated. Strides were selected for analysis on the basis of having a velocity in the range of 6.0-7.0 m/s, and multivariate analysis of variance was used to detect significant differences in the stride kinematics of horses trained for the two sports (p < or = 0.01). The average velocity of the dressage horses was 6.37 m/s, compared with 6.40 m/s for the racehorses. There were no significant differences between the two groups in velocity, stride duration, stride length or the distances between limb placements. The stance durations of all four limbs and the overlaps between them were longer, whereas the duration of the suspension phase was shorter in the dressage horses than in the racehorses (p < or = 0.01). The time between impacts of the diagonal limb pair was close to zero in both groups, with individual horses showing some variability in the order of placement of the diagonal limb pair. However, the sequence of footfalls was not significantly different between the two groups (p < or = 0.01).