Directional preferences have previously been demonstrated during horizontal arm movements. These preferences were characterized by a tendency to exploit interaction torques for movement production at the shoulder or elbow, indicating that the preferred directions depend on biomechanical, and not on visual perception-based factors. We directly tested this hypothesis by systematically dissociating visual information from arm biomechanics. Sixteen subjects performed a free-stroke drawing task that required performance of fast strokes from the circle center toward the perimeter, while selecting stroke directions in a random order. Hand position was represented by a cursor displayed in the movement plane. The free-stroke drawing was performed twice, before and after visuomotor adaptation to a 30° clockwise rotation of the perceived hand path. The adaptation was achieved during practicing pointing movements to eight center-out targets. Directional preferences during performance of the free-stroke drawing task were revealed in ten out of the sixteen subjects. The orientation and strength of these preferences were largely the same in both conditions, showing no significant effect of the visuomotor adaptation. In both conditions, the major preferred directions were characterized by higher contribution of interaction torque to net torque at the shoulder as well as by relatively low inertial resistance and the sum of squared shoulder and elbow muscle torques. These results support the hypothesis that directional preferences are largely determined by biomechanical factors. However, this biomechanical effect can decrease or even disappear in some subjects when movements are performed in special conditions, such as the virtual environment used here.