Brain-machine interface (BMI) systems hold the potential to return lost functions to patients with motor disorders. To date, most efforts in BMI have concentrated on decoding neural activity from forearm areas of cortex to operate a robotic arm or perform other manipulation tasks. Efforts have neglected the locomotion functions of hindlimb/trunk cortex. However, the role of cortex in hindlimb locomotion of intact rats, which are often model systems for BMI testing, is usually considered to be small. Thus, the quality of representations of locomotion available in this area was uncertain. We designed a new rodent BMI system, and tested decoding of the kinematics of trunk and hindlimbs during locomotion using linear regression. Recordings were made from the motor cortex of the hindlimb/trunk area in rats using arrays of six tetrodes (24 channels total). We found that multiple movement-related variables could be decoded simultaneously during locomotion, ranging from the proximal robot/pelvis attachment point, and the distal toe position, through hindlimb joint angles and limb endpoint in a polar coordinate system. Remarkably, the best reconstructed motion parameters were the more proximal kinematics, which might relate to global task variables. The pelvis motion was significantly better reconstructed than any other motion features.