Torsion tests have been used to evaluate the shear response of concrete. The complete load-deformation response of unnotched concrete specimens subjected to torsion is difficult to obtain because of test instabilities. In this paper torsional tests with a closed-loop test system are performed on hollow concrete cylinders. The test instability is shown to be a result of snapback in the postpeak part of the torque-twist response of these specimens. The failure is a result of the propagation of a single major crack. A test signal that is a combination of torque and gauge rotation is introduced and is demonstrated to give improved control by eliminating the test instability. The response of concrete specimens subjected to both controlled and uncontrolled torsion tests, using specially designed test equipment, are presented and analyzed in terms of propagation of a single major crack.