Evidence suggests that some 10 m.y. ago a large portion of thickened mantle lithosphere below the Tibetan Plateau was rapidly removed. Removal appears to have been abrupt and delayed from the time of substantial thickening. Fluid dynamical models of convective thermal boundary layer instability have had difficulties explaining abrupt lithospheric removal long after initial thickening. Such models have implied that either removal should be synchronous with thickening or should not occur at all, due to boundary layer stabilization resulting from high lithospheric viscosity. We present simple thermal/chemical boundary layer convection models that suggest a compromise between previous end-member results. The models allow for delayed instability of a thickened thermal boundary layer under geologically reasonable parameter conditions. Instability delay relies on high viscosity of thickened mantle lithosphere stabilizing it against free convective removal and on lateral extent of thickened crust exceeding that of mantle lithosphere. The latter condition leads to a peripheral region of thickened crust and relatively unthickened mantle lithosphere and an interior plateau region of both thickened crust and mantle lithosphere. If the thermal resistance and/or the rate of internal heat production of crust is greater than that of mantle, then mantle lithosphere at the peripheral region can be thermally eroded to the point that the base of the crust comes into contact with bulk mantle. This creates a peripheral weak zone that allows thickened mantle lithosphere of the interior plateau region to become mechanically detached and sink into bulk mantle. The instability is akin to delamination, as originally defined, and its delay time is roughly the thermal diffusion time across the thinnest portion of mantle lithosphere. ¿ American Geophysical Union 1995