The torsional split Hopkinson bar (SHB) is an important method to study the dynamic shear behaviour and shear localization of materials under high strain rates. Different specimen sizes were used in literatures, and the size of the specimen might have an effect on the experimental results. Numerical simulation on torsional SHB tests was carried out with LS-DYNA. The strain signal on the incident and transmitted bars were obtained from the simulation just as the experiment. Then the numerical strain-stress relationship of the material was derived from the numerical strain signal using the experiments data process of torsional SHB. The agreement between numerically derived strain-stress results and the specimen material properties specified in numerical modelling indicates that the torsional SHB is applicable to study the dynamic shear behaviour of materials under high strain rates. The specimen gauge diameter has no significant effect on the dynamic torsional test result. However, higher adhesive strength is required to fix the larger gauge diameter specimen on the bars. The specimen gauge thickness has little effect on the experimental results with a modified formula to calculate the specimen stress. Still, the increase of specimen gauge thickness will lead to the increase of non-uniformity of specimen stress and strain (strain rate). Based on the simulation analysis, suggestions on the specimen size design are given as well.