An experiment investigated the ability by human observers to detect temporal reversals in dynamic displays of human locomotion. We video-taped the lower portion of the body of actors walking at their preferred speed either in the normal, forward direction (FW) or in the backward direction (BW). The videos were presented in a random order either as recorded (N) or in reverse (R). In one session, we presented both normal and time-reversed stimuli in the original upright orientation. In a second session, the stimuli were rotated by 180° around the horizontal axis. Observers were informed that the real recorded movement was either forward or backward and were asked to decide whether or not the movement had been time-reversed prior to the presentation. Although the kinematics of forward and backward human locomotion is quite similar, the detection of temporal reversals followed a consistent pattern showing a good accuracy in condition FW-N and a reduced but still above-chance performance in condition BW-R (by design, in both conditions actors appeared to walk forward). Performance was instead at chance level in the other two conditions where the apparent direction of the movement was backward. Inverting the spatial orientation of the stimuli reduced but did not suppress the ability to detect temporal reversals in the two conditions with apparent forward direction of movement. It is argued that implicit motor competence is at least in part instrumental for extracting the subtle discriminal information from the stimuli.