An adaptive method for detached-eddy simulation (DES) is tested by simulations of flow in a family of three-dimensional (3D) diffusers. The adaptive method either adjusts the model constant or defaults to a bound if the grid is too coarse. On the present grids, the adaptive method adjusts the model constant over most of the flow, without resorting to the default. Data for the diffuser family were created by wall-resolved, large-eddy simulation (LES), using the dynamic Smagorinsky model, for the purpose of testing turbulence models. The family is a parameterized set of geometries that allows one to test whether the pattern of separation is moving correctly from the top to the side wall as the parameter increases. The adaptive DES model is quite accurate in this regard. It is found to predict the mean velocity accurately, but the pressure coefficient is underpredicted. The latter is due to the onset of separation being slightly earlier in the DES than in the LES.