Recordings were made from single ganglion cells in the isolated retinas of goldfish. Statistics of the maintained discharge were computed for steady-state firing in the dark, in the presence of a steady-adapting light confined essentially to the center of the receptive field, and in the presence of the same adapting light modulated by Gaussian white noise. 2. Steady adapting lights effected a significant initial change in firing rate and affected sensitivities to test probes. After a new minutes, however, the rate returned to a moderate level. There was no consistent relationship between firing in the dark and illuminated conditions, nor was there a consistent relationship between the values of the coefficient of variation of firing in the dark and in light, although some individual cells showed changes in these properties when illuminated. 3. There was no consistent change in the organization of the firing in dark and in steady light. Neither the normalized autocovariances nor the serial correlograms, indicators of short-term order, were significantly affected by light. Similarly, plots of the standard deviation of rate versus sample duration, an indicator of long-term structure, were not altered by illumination. The intrinsic noise thus appears unaffected by retinal illumination. 4. Gaussian modulation of an adapting light (extrinsic noise) increased the coefficient of variation; it also dramatically increased mean firing rate. This effect was observed in both red-on and red-off cells; it indicates a nonlinearity that is present even in the spatially linear X-like neurons. 5. Gaussian modulation of an adapting light made the short-duration values of the normalized autocovariances slightly more positive. Similarly, modulation made the first two or three coefficients of the serial correlogram more positive; this indicates low pass filtering of the extrinsic noise. 6. Gaussian modulation of an adapting light had no effect on the plots of standard deviation of rate versus sample duration other than simple translation due to increased variability. Since the form of these plots in the absence of modulation implies high-pass filtering of the intrinsic noise, the extrinsic noise must also be processed by the same (or a similar) high-pass filter. 7. A model is proposed in which extrinsic signals are band pass filtered by a low-pass and a high-pass filter in cascade. The intrinsic variability evident in the maintained discharge is apparently injected between the two filters. Alternative possibilities are discussed.
