The hitherto particularly obscure process of deformation of a single crystal by torsion is elucidated in the general case where the slip plane is inclined to the torsion axis. It is shown to consist primarily, in zinc crystals about 0$\cdot $8 cm in diameter, of a new type of slip which will be called 'flexural' rotational slip. In this slip process the slip lamellae are rotated relative to adjacent lamellae, about the common normal at any point of the interface, but with simultaneous development of pronounced saddle-like flexure. This flexure results in a variation in the amount of relative rotation of neighbouring lamellae over their interface, and also causes a cylindrical crystal to develop a rounded-triangular cross-section. Optical examination shows S-shaped (0001) slip bands about 0$\cdot $05 to 0$\cdot $1 mm wide. Electron diffraction shows that at the saddle-shaped (0001) cleavage face the lattice is mainly practically undistorted (apart from the macroscopic flexure) to a depth of 200 angstrom or more, thus the slip lamellae are at least 200 angstrom thick. An expression defining the shape the slip lamellae must take up is deduced assuming that the elastic deformation is negligible, and this gives good quantitative agreement with most regions of the observed (0001) surfaces. The very minor role of twinning in the range examined (up to 66 degrees /cm torsion) is demonstrated by this agreement of the calculated and observed (0001) surface shape. Experiments on torsion of a model made of a hundred lead lamellae illustrate well the development of the saddle-like flexure, the rounded-triangular rod section, and the non-uniformity of rotation of the lamellae over their interface.