Colour constancy in fishes was shown to exist already at the retinal level, and horizontal cells (HCs) were supposed to take part in it [1-3]. HCs collect signals from a wide area and might supply the visual system with the information about ambient illumination to make necessary scaling for the output cone signals by means of some feedback mechanism.
          Unlike the feedforward version, such a scheme of discounting the illumination meets with some difficulties when realized as a feedback circuit.
          First, the point is that without an additional amplification in the loop the feedback fails to fulfil good colour constancy (see p. 5-6).
          Second, let us assume that the feedback from horizontal cells does perform perfect colour constancy in some way. Then we meet the following apparent contradiction. A perfect colour constancy means that cone outputs (glutamate released in the cone synapse) should be already invariant to the level of illumination. Apparently in such case HCs do not receive any information about the illumination also and cannot provide the necessary feedback.
The contradiction becomes even more evident in case of uniform stimulation of the retina, when HCs responses must depend on the stimulus intensity, while the glutamate release does not.