We used three-dimensional coordinate data for six points recorded from six people picking up, drinking from and replacing a glass on a table. We recorded twelve repetitions for each person. There was one point each on the right shoulder, elbow, wrist, and temple and two points on the hand. The average duration of the sequences was 7.5 s with a standard deviation of 1.1 s. Position was recorded from infra red emitters attached at these points using a 3-D position analysis system, the Optotrak from Northern Digital. Spatial resolution was of the order of a millimeter and temporal resolution 60 Hz.

The three-dimensional co-ordinate data was imported to Matlab and used to generate two bit motion point light images with 3 x 3 pixel white squares indicating the position of each point on a black background. Images showed a sagittal view of the actors facing right and were used to produce a QuickTime animation at 30 Hz. All actors were right handed. Animations were presented on a Power Macintosh 7600/132. Animations were spatially normalized by scaling so that they all had the same maximum spatial extent. Thus scaled the images measured 10 cm x 10 cm. Viewed from a distance of approximately 60 cm this gave a viewing angle of 9.5°. The rest of the screen was black.

The tangential velocity of the wrist was calculated for each frame of each sequence and plotted as a velocity profile of the movement (see Figure 2 a). Key frames separating the phases of movement were defined on the wrist velocity profile. All except one corresponded to minima of velocity between different phases of movement. The remaining time point corresponded to a point on the velocity profile where there was a marked change in acceleration and which corresponded to the onset of drinking. These points were defined automatically initially but were then checked and edited by eye. Six time points were defined as shown in Figure 2 a) and corresponded to the start of movement, picking up the glass, the onset and offset of drinking putting down the glass and returning to the initial position.

Grand average values for each of the six key frames were calculated from the twelve repetitions of all six individuals, a total of seventy-two sequences. For each individual sequence the difference of its key frames from the average values were calculated. A proportion of this difference was then added to the original value. This had the effect of exaggerating the durations of the segments between key frames. The proportions used were -0.5, 0, 0.5 and 1. For time normalized sequences key frame values were scaled so that their overall duration was the average value for the actor.

The new key frames were then used to resample the original motion data and generate the required number of frames for the exaggerated sequence. Linear interpolation was used to generate intermediate frames from the original data. Thus temporal properties were changed by increasing or decreasing the number of frames, not by altering frame duration or inter stimulus intervals, manipulations known to affect the processing of biological motion (Mather, Radford & West, 1992; Thornton et al 1998). QuickTime animations of temporally exaggerated sequences were produced in the same manner as for the unexaggerated sequences.