We tested the hypothesis that dominant and nondominant overarm throws of different speeds are made by time-scaling of joint rotations, i.e., by joint rotations which have the same positions and amplitudes but which are scaled in time. Eight skilled subjects stood and made overarm throws with both their dominant and nondominant arms. Six joint rotations were computed from recordings of arm segments made with the search-coil technique. Throws made with nondominant arms were less accurate and had lower ball speeds. In contrast to the hypothesis, dominant arms showed large and consistent differences between fast and slow throws in 6-dimensional angular position joint space. These same throws showed similar hand angular paths when these were time-scaled based on ball speed. Nondominant arms showed only small differences in angular position joint space in fast and slow throws. It is concluded that a joint space pattern resembling that predicted by time-scaling occurs in nondominant arm throwing when it is unskilled. However, time-scaling does not occur in dominant arm throwing, i.e., a skilled fast throw is not simply a skilled slow throw whose joint positions and amplitudes remain constant but whose joint velocities are sped-up. We hypothesize for future study that when subjects first learn to throw at different speeds with their dominant arms they use time-scaling of joint rotations which involves compensating for interaction torques; then as they become skilled at throwing fast, time-scaling is superseded by a more complex pattern of interjoint coordination which involves exploiting interaction torques.