Then the relation (13) can be used to express the moments mx(r niy(r and mxy(r^ in terms of normal moments of the triangular element. Since the normal moments m?, Wg, and mg are in no way coupled to unknowns of an adjacent element (see Fig. lb), it is possible to eliminate these unknowns per element using Gauss elimination. This reduces the final number of degrees of freedom from 15 to 12, which is identical to the number of degrees of freedom in the completely compatible, displacement type element of Clough and Felippa.1 3. Discussion of Results In Fig. 2, four different plate bending problems are drawn. The square plate is either simply supported (ss) or clamped (c) and it is loaded by a uniformly distributed load (u) or by a concentrated central force (c). For these four cases, the central deflections and four moments are determined. The results obtained with the Herrmann element and with the refined element are compared with results given by Timoshenko.4 Only one octal of the square plate is needed to derive the results for these problems. The errors (in per cents) are given in Table 1; here n denotes the number of elements in which half a side of the square is divided. An error of zero is an error of less than 0.5%; a plus or minus stands for overor underestimation of a certain quantity. The average errors in the four deflections and four moments show that a division of half a side in eight Herrmann elements leads to about the same accuracy as a division in two parabolic elements. Particularly, it can be observed that the Herrmann element gives accurate values of the moments in the center of the plate where dM fbx = 0, but the errors in the moments along the edge are large. This phenomenon, that the accuracy of displacements obtained with the Herrmann element depends on the value of dTkf/dx, and not on M, can also be observed in simple beam bending problems. The accuracy of the central deflection of the plates using the refined element is about the same as the accuracy that can be obtained with the same