An orientational difference of only 0.3-0.5 deg can be discriminated between two gratings or two lines, although psychophysical channels and cortical cells both have comparatively broad orientation bandwidths of 10-25 deg. One proposed explanation for the fineness of orientation discrimination is that, while detection is determined by the most excited orientation-tuned neural elements, superthreshold orientation discrimination is determined by difference signals between these elements [Westheimer et al., J. Opt. Soc. Am. 66, 332 (1976)]. This implies that, if stimulus orientation is changed slightly, the most important elements for discriminating this change will be those whose relative activity changes most, even though the excitation of these elements may be comparatively weak. In accord with this prediction, we found that adapting to a high-contrast grating degraded discrimination for test gratings inclined at about 10-20 deg to the adapting grating while having little effect on the detection of these inclined gratings. For test gratings parallel to the adapting grating, discrimination was improved, but detection was degraded. Either an opponent-process or a line-element model can account for these effects of adaptation. An opponent model can also explain our findings that subjects do not confound orientation change with contrast change and that suprathreshold orientation discrimination is almost independent of contrast, varying by only +/- 10% from about 3 to about 25 times contrast threshold. A discrimination model must incorporate reliable storage of spatial frequency, because discrimination was not affected by increasing the interval between grating presentations from 1 to 10 sec.(ABSTRACT TRUNCATED AT 250 WORDS)