Whole-arm motor tasks performed by nonhuman primates have become a popular paradigm to examine neural activity during motor action, but such studies have traditionally related cell discharge to hand-based variables. We have developed a new robotic device that allows the mechanics of the shoulder and elbow joints to be manipulated independently. This device was used in the present study to examine neural activity in primary motor cortex (MI) in monkeys (Macaca mulatta) actively maintaining their hand at a central target as they compensated for loads applied to the shoulder and/or elbow. Roughly equal numbers of neurons were sensitive to mechanical loads only at the shoulder, only at the elbow, or loads at both joints. Neurons possessed two important properties. First, cell activity during multi-joint loads could be predicted from its activity during single-joint loads as a vector sum in a space defined by orthogonal axes for the shoulder and elbow. Second, most neurons were related to flexor torque at one joint coupled with extensor torque at the other, a distribution that paralleled the observed activity of forelimb muscles. These results illustrate that while MI activity may be described by independent axes representing each mechanical degree-of-freedom, neural activity is also strongly influenced by the specific motor patterns used to perform a given task.