Marchand and Duffy tested thin-walled steel tubes in a split Hopkinson torsion bar at a nominal strain-rate of approximately 1600/s and could not determine conclusively whether a shear band initiating at a point in the tube propagated around the circumference in one direction or in both directions. They estimated the speed of propagation to be 520 m/s in the former case and 260 m/s in the latter. Here we simulate their test numerically, and find that the shear band propagates in both directions around the circumference of the tube. When the tube is twisted at a nominal strain-rate of 5000/s, the band speed varies from 180 m/s at the site of the initiation to approximately 1000 m/s at the nearly diametrically opposite point. The band speed increases with an increase in the nominal strain-rate. The material defect is modeled by assuming that a small region near the center of the tubular surface is made of a material weaker than that of the rest of the tube.