SUMMARYA number of teleost fishes have photoreceptor mechanisms to detect linearly polarized light. We studied the neuronal mechanism underlying this ability. It was found that a polarized signal could be detected in rainbow trout (Oncorhynchus mykiss) both in the electroretinogram (ERG) and in the compound action potential (CAP) measured in the optic nerve, indicating a strong retinal contribution to the processing of polarized light. The CAP recordings showed a W-shaped sensitivity curve, with a peak at 0°, 90°a nd 180°, consistent with processes for both vertical and horizontal orientation. By contrast, the ERG recordings reveal a more complex pattern. In addition to the peaks at 0°, 90° and 180°, two additional peaks appeared at 45° and 135°. This result suggests a specialized contribution of the outer retina in the processing of polarized light. The spectral sensitivity of the mechanisms responsible for these intermediate peaks was studied using chromatic adaptation. Here we show that long wavelength-sensitive (LWS) cone mechanism adaptation shifted the intermediate peaks towards 90°, whereas ultraviolet-sensitive (UVS) cone mechanism adaptation shifted the peaks away from 90° towards either 0° or 180°. These results provide further confirmation that the 90° peak is dominated by the LWS cone mechanism and the 0° and 180° peaks are dominated by the UVS cone mechanism. In addition, a pharmacological approach was used to examine the retinal neural mechanisms underlying polarization sensitivity. The effect of blocking negative feedback from horizontal cells to cones on the ERG was studied by making intraocular injections of low doses of cobalt, known to block this feedback pathway. It was found that the intermediate peaks seen in the ERG polarization sensitivity curves were eliminated after application of cobalt, suggesting that these peaks are due to outer retinal inhibition derived from feedback of horizontal cells onto cones. A simple computational model was developed to evaluate these results. The model consists of opponent and non-opponent processing elements for the two polarization detectors. This model provides a first approximation analysis suggesting that opponent processing occurs in the outer retina for polarization vision. Although it seems that polarization vision uses a slightly more complicated coding scheme than colour vision, the results presented in this paper suggest that opponent and non-opponent channels process polarization information.