This paper deals with the notion of disturbance model uncertainty. The disturbance is modeled as the output of a first-order filter which is driven by white noise and whose bandwidth and gain are uncertain. An analytical expression for the steady-state output variance as a function of the uncertain bandwidth and gain is derived, and several properties of this variance function are analyzed. Two notions, those of disturbance bandwidth margin and disturbance gain margin are also introduced. These tools are then applied to the analysis of a simple altitude-hold autopilot system in the presence of turbulence where the turbulence scale is treated as an uncertain parameter. It is shown that the autopilot, which is satisfactory for nominal turbulence scale, may be inadequate when the uncertainty is taken into account. Moreover, it is proven that, in order to obtain a design that provides robust performance in the face of turbulence scale uncertainty, it is necessary to substantially increase the controller bandwidth, even if one is willing to sacrifice the autopilot's holding ability and stability robustness.