Why are some spatial discriminations independent of contrast?

The accuracy of some spatial discriminations, including spatial frequency and orientation, is independent of the contrasts of the stimuli discriminated except when contrasts are near the detection threshold. This fact is surprising because higher contrasts should improve signal-to-noise ratios and performance. Two alternative explanations of this puzzle are examined, using a signal-detection-theory vector model: (1) noise increases with stimulus contrast in such a way that signal-to-noise ratios remain constant; (2) noise is constant, but the difference signal generated by each individual tuned mechanism is independent of contrast because the response function of the mechanism becomes compressive (approximately logarithmic) at a low contrast. The alternative explanations generate different predictions, which have been tested in several experiments. The results reject alternative (1) as a model of suprathreshold discrimination and give mixed support to alternative (2).