Contrast threshold elevation was measured in human observers at a single spatial frequency (4 or 8 c/deg) after adaptation to gratings at different contrasts and spatial frequencies. When plotted against adapting contrast, the threshold elevation functions for different adapting frequencies were neither straight nor parallel, which poses difficulties for the "equivalent contrast transformation". The different functions appear to form a single family, differing by a scaling factor that depends on spatial frequency but is independent of contrast. This is similar to expressing the aftereffect as a proportion of its peak value. A multiple channel, "fatigue" model of adaptation is shown to be consistent with the results, provided the channel shape is appropriate and channels overlap considerably in the frequency domain. Channel bandwidth is estimated at 1.4 octaves, similar to that of cortical cells.