Numerous studies have examined control of force magnitude, but relatively little research has considered force direction control. In this study, subjects applied isometric forces to a handle and we compared within-trial variability when producing force in different directions. The standard deviation (SD) of the force parallel to the prescribed direction of force production increased linearly with the targeted force level, as did the SD of the force perpendicular to the instructed direction. In contrast, the SD of the angle of force production decreased with increased force level. In the four (of eight) instructed force directions where the endpoint force was generated due to a joint torque in only one joint (either the shoulder or elbow) the principal component axes in force space were well aligned with the prescribed direction of force production. In the other directions, the variance was approximately equal along the two force axes. The variance explained by the first principal component was significantly larger in torque space compared to the force space, and mostly corresponded to positive correlation between the joint torques. Such coordinated changes suggest that the torque variability was mainly due to the variability of the common drive to the muscles serving two joints, although this statement needs to be supported by direct studies of muscle activation in the future.