The motion of continents relative to the Earth's spin axis may be due either to rotation of the entire Earth relative to its spin axis--true polar wander--or to the motion of individual plates. In order to distinguish between these over the past 320 Myr (since the formation of the Pangaea supercontinent), we present here computations of the global average of continental motion and rotation through time in a palaeomagnetic reference frame. Two components are identified: a steady northward motion and, during certain time intervals, clockwise and anticlockwise rotations, interpreted as evidence for true polar wander. We find approximately 18 degrees anticlockwise rotation about 250-220 Myr ago and the same amount of clockwise rotation about 195-145 Myr ago. In both cases the rotation axis is located at about 10-20 degrees W, 0 degrees N, near the site that became the North American-South American-African triple junction at the break-up of Pangaea. This was followed by approximately 10 degrees clockwise rotation about 145-135 Myr ago, followed again by the same amount of anticlockwise rotation about 110-100 Myr ago, with a rotation axis in both cases approximately 25-50 degrees E in the reconstructed area of North Africa and Arabia. These rotation axes mark the maxima of the degree-two non-hydrostatic geoid during those time intervals, and the fact that the overall net rotation since 320 Myr ago is nearly zero is an indication of long-term stability of the degree-two geoid and related mantle structure. We propose a new reference frame, based on palaeomagnetism, but corrected for the true polar wander identified in this study, appropriate for relating surface to deep mantle processes from 320 Myr ago until hotspot tracks can be used (about 130 Myr ago).