Strategies used by the central nervous system (CNS) to optimize arm movements in terms of speed, accuracy, and resistance to fatigue remain largely unknown. A hypothesis is investigated here that the CNS exploits biomechanical properties of multijoint limbs to increase efficiency of movement control. To test this notion, a novel free-stroke drawing task was employed that instructs subjects to make straight strokes in as many different directions as possible in the horizontal plane via rotations of the elbow and shoulder joints. In spite of explicit instructions to distribute strokes uniformly, subjects demonstrated biases to move in specific directions. These biases were associated with a tendency to perform movements that included active motion at one joint and largely passive motion at the other joint, revealing a tendency to minimize intervention of muscle torque for regulation of the effect of interaction torque. Other biomechanical factors, such as inertial resistance and kinematic manipulability, were unable to adequately account for these significant biases. Also, minimizations of jerk, muscle torque change, and sum of squared muscle torque were analyzed; however these cost functions failed to explain the observed directional biases. Collectively, these results suggest that knowledge of biomechanical cost functions regarding IT regulation is available to the control system. This knowledge may be used to evaluate potential movements and to select movement of "low cost". The preference to reduce active regulation of interaction torque suggests that, in addition to muscle energy, the criterion for movement cost may include neural activity required for movement control.