The purpose of this paper was to describe a technique that enables three-dimensional (3D) gait kinematics to be obtained using an electromagnetic tracking system, and to report the intra-trial, intra-day/inter-tester and inter-day/intra-tester repeatability of kinematic gait data obtained using this technique. Ten able-bodied adults underwent four gait assessments; the same two testers tested each subject independently on two different days. Gait assessments were conducted on a custom-built long-bed treadmill with no metal components between the rollers. Each gait assessment involved familiarisation to treadmill walking, subject anatomical and functional calibration, and a period of steady-state treadmill walking at a self-selected speed. Following data collection, 3D joint kinematics were calculated using the joint coordinate system approach. 3D joint angle waveforms for 10 left and right strides were extracted and temporally normalised for each trial. Intra-trial, intra-day/inter-tester and inter-day/intra-tester repeatability of the temporally normalised kinematic waveforms were quantified using the coefficient of multiple determination (CMD). CMDs for joint kinematics averaged 0.942 intra-trial, 0.849 intra-day/inter-tester and 0.773 inter-day/intra-tester. In general, sagittal plane kinematics were more repeatable than frontal or transverse plane kinematics, and kinematics at the hip were more repeatable than at the knee or ankle. The level of repeatability of kinematic gait data obtained during treadmill walking using this protocol was equal or superior to that reported previously for overground walking using image-based protocols.