Goal: to partition the gain changes in the visual system into two serial stages and to measure their relative contribution over a 10,000-fold change of illumination. Rationale and method. Superimposing on the retina two interference fringes of different spatial frequency or orientation (MacLeod, Williams, & Makous, VR, 1992) creates an illusory grating (a distortion product) at a nonlinear stage that probably lies in the outer retina (Chen, Makous, & Williams, VR, 1993). The amplitude of the signal representing a distortion grating is proportional to the product of the amplitudes of the signals representing the two fringes at the nonlinear stage: hence the distortion signal changes in proportion to the square of any gain changes preceding the nonlinearity, but that of a single fringe grating changes only in direct proportion to the gain changes; both kinds of signal change equally with any gain changes following the nonlinear stage. By comparing sensitivities to fringe and distortion gratings (in 3 observers) one can infer how much of the gain changes precede and follow the nonlinear stage. Results. (1) Weber's law holds within 1% for 500 msec fringe gratings presented against steady backgrounds, and also for 2 msec fringes flashed in the dark. (2) As Troland value increases from 10 to 1,000 Td, sensitivity to a distortion product decreases 3- to 5-fold relative to that of a fringe grating of the same spatial frequency (10 cpd), but the relative sensitivity does not change outside that range. (3) The gain governing sensitivity to 2 msec fringes is completely determined by the energy in the test flash: steady backgrounds from 1 to 1,000 Td have no detectable effect on sensitivity to a 3,300 Td fringe. Conclusions: (1) as total sensitivity varies over a 10,000-fold range, the gain preceding the nonlinear stage is no more than 5-fold; and (2) a 2 msec flash completely predominates over weaker, steady fields in setting gain.