We compared image computation in the rabbit retina by two different cell types: the so-called ‘local edge detecting’ ganglion cells and the well-known brisk–sustained ganglion cells. From both anatomical and physiological evidence, these cells are present in nearly equal numbers and thus overlap to sample the same regions of visual space. We recorded simultaneously from overlapping cells on a dense microelectrode array. The results were analysed using an anatomically realistic simulation of the retina's processing levels. The ‘local edge detecting’ cell was found to be tuned to higher spatial frequencies and to have a narrower spatial frequency bandpass than the brisk–sustained cells. Simulation revealed that this is due primarily to the ‘zero-crossing’ detector implied by the definition of the local edge detector. The outputs of the simulations in response to complex images were analysed quantitatively. The results showed the population of local edge detectors to transmit a sparser code than the brisk–sustained cells.