It is generally accepted that each cortical hemisphere primarily drives the opposite side of the body. Yet, primary motor cortical (M1) activity has been robustly correlated with both contralateral and ipsilateral arm movements. It has remained unanswered as to why ipsilaterallyrelated activity does not cause contralateral motor activity. Here we apply multi-joint elbow and shoulder loads to the left or right arms of monkeys during a postural perturbation task. We show that many M1 neurons respond to mechanical disturbances applied to either the contra-or ipsilateral arms. More neurons respond to loads applied to the contralateral arm with response magnitudes that were ~2x as large and had onset times that were ~10ms earlier. However, in some cases, neurons exhibited large and earlier responses to loads applied to the ipsilateral arm than loads applied to the contralateral arm. Similar effects were observed when the monkeys were maintaining postural control well after the load had been applied. Importantly, we show that the load preference to one arm has little predictive power on a neuron's preference in the opposite arm. Furthermore, we found contralateral and ipsilateral neural activity resided in orthogonal subspaces allowing for a weighted sum of neural responses to extract the contralateral activity without interference from the ipsilateral activity, and vice versa. These data show how activity in M1 unrelated to downstream motor targets can be segregated from downstream motor output.the torques were 0.20 Nm for single joint loads, and 0.14 Nm at each joint for multi-joint torques. All torque conditions were completed in random order, comprising one block. A minimum of 10 blocks were completed.Neural, EMG and kinematic recordings. After training, monkeys underwent surgery to implant 96-channel Utah Arrays (Blackrock Microsystems, Salt Lake City, UT) into the arm region of M1. Surgeries were performed under aseptic conditions and a head fixation post was also attached to the skull using dental cement. Spike waveforms were sampled at 30kHz and acquired using a 128-Channel Neural Signal Processor (Blackrock Microsystems, Salt Lake City, UT). Spikes were manually sorted offline (Offline Sorter, Plexon Inc., Dallas TX) using a space spanned by the top two principal components and the peak-trough voltage difference. Only well isolated single-units were used for analysis. For Monkey P we also implanted a chamber above the right arm area of M1 and neurons were recorded via single electrode over multiple recording sessions using standard techniques (Herter et al., 2009).We recorded intramuscular electromyographic (EMG) activity from Monkey P by inserting two thin wires into the muscle belly of brachioradialis, lateral and long head of triceps, long head of the biceps and pectoralis major. Signals were digitized at 1KHz. Joint angles, velocities, and accelerations for both arms, were recorded at 1 kHz by the Neural Signal Processor. All offline analysis was performed using custom MATLAB scripts (The MathWorks, Inc., ...