Differentiation of nucleated erythrocytes involves transformation from spheroids to flattened discoids to mature flattened ellipsoids. The marginal band (MB) of microtubules is required for this process and continues to play a role in maintaining mature ellipsoidal cell shape. One hypothesis for MB function is that cell ellipticity is generated and maintained by asymmetric application of force across a flexible, circular MB frame by the membrane skeleton or other transverse elements. This is based on an earlier finding that isolated erythrocyte MBs are much more circular than MBs in situ. However, our present studies of salamander erythrocyte MBs isolated by a detergent-based method challenge this hypothesis. Most of these isolated MBs are initially elliptical, even though they lack transverse material (= E-MBs). They can be stabilized in that form for long periods and can be converted experimentally into the circular form (= C-MBs) by extended incubation in isolation medium or by treatment with elastase or subtilisin. We have tested an alternative hypothesis for generation and maintenance of ellipsoidal MBs, one based on intrinsic differential bending resistance and supported by construction of models. Using laser microsurgical transection to compare mechanical responses of isolated E-MBs and C-MBs, we have found their behavior to be quite different. Whereas C-MBs linearize, most E-MBs do not, instead retaining considerable curvature. These results are incompatible with the differential bending resistance hypothesis, which predicts both C-MB and E-MB linearization. However, they are consistent with a third model, in which material bound to the MB stabilizes it in the mature ellipsoidal form, and indicate that the mechanism for maintenance of MB ellipticity differs from that involved in its generation.