Brain-machine interface (BMI) efforts have been focused on using either invasive implanted electrodes or training-extensive conscious manipulation of brain rhythms to control prosthetic devices. Here we demonstrate an excellent prediction of movement trajectory by real-time magnetoencephalography (MEG). Ten human subjects copied a pentagon for 45 s using an X-Y joystick while MEG signals were being recorded from 248 sensors. A linear summation of weighted contributions of the MEG signals yielded a predicted movement trajectory of high congruence to the actual trajectory (median correlation coefficient: r=0.91 and 0.97 for unsmoothed and smoothed predictions, respectively). This congruence was robust since it remained high in cross-validation analyses (based on the first half of data to predict the second half; median correlation coefficient: r=0.76 and 0.85 for unsmoothed and smoothed predictions, respectively).