According to simple theory, torque measurements on cubic single‐crystal disks should give values for the crystal anisotropy constants,K1andK2, that are independent of field above some minimum saturating field. Our experiments on {100} and {110} disks of 3¼% silicon‐iron, and previous experiments of others, show that this theory is inadequate. In a more refined theory it is assumed that the edges of the disk are not saturated even in high fields. The observed increase in the peak values of the torque with increasing field can be attributed to a very small variation of the net magnetizationIas the disk is rotated in the field. This variation must be such thatIis largest in the hard directions of magnetization and smallest in the easy directions. Direct observation of the domain patterns on a {100} disk appears to confirm these assumptions. For both {100} and {110} disks,K1′ (the effective value ofK1) varies as 1−c/√Hover the range 1000 to 20 000 oersteds. The correspondingK2′ values for the {110} disk show a more complicated dependence onH. Values ofK1′ found from torque peaks differ somewhat from those obtained from torque slopes; the difference between the smallerK2′ values is much more serious. The discrepancy is attributed to the particular way in whichIvaries with disk rotation.