The purpose for conducting a torsional study of a proposed design is to predict torsional natural frequencies and torsional responses due to dynamic torques associated with the driven and/or driving equipment. If the mean, or constant, response plus that due to torsional vibration is predicted to exceed a safe level, then changes are proposed to ensure reliable performance. The validity of the predictions is strongly dependent upon the accuracy of the mass-elastic properties that are used in the model. The modeling technique should account for the fact that there is always some level of uncertainty about these properties. A closely related concern is that the actual installation may differ from the drawings and other details supplied for the study; some such deviations can significantly affect the torsional response. This paper discusses these effects and describes a methodology for handling the uncertainties. Typical levels of uncertainty in predicted torsional natural frequencies are presented. Field results are compared to predictions. Real world examples show what can happen when the model and the machine differ in torsional properties.