The synaptic conductance of the On-Off direction-selective ganglion cells was measured during visual stimulation to determine whether the direction selectivity is a property of the circuitry presynaptic to the ganglion cells or is generated by postsynaptic interaction of excitatory and inhibitory inputs. Three synaptic asymmetries were identified that contribute to the generation of direction-selective responses: (1) a presynaptic mechanism producing stronger excitation in the preferred direction, (2) a presynaptic mechanism producing stronger inhibition in the opposite direction, and (3) postsynaptic interaction of excitation with spatially offset inhibition. Although the on-and off-responses showed the same directional tuning, the off-response was generated by all three mechanisms, whereas the on-response was generated primarily by the two presynaptic mechanisms. The results indicate that, within a single neuron, different strategies are used within distinct dendritic arbors to accomplish the same neural computation.Key words: direction selectivity; ganglion cells; synaptic conductance; inhibition; excitation; dendritic integration; on-and off-pathways; rabbit retinaThe direction-selective ganglion cells (DSGCs) in the rabbit retina are a model system for investigating neural computation (Vaney et al., 2001). These cells respond strongly to an image moving in a preferred direction but only weakly to an image moving in the opposite "null" direction. The foundation for understanding the cellular mechanisms of direction selectivity in vertebrates was laid by Barlow and Levick (1965), whose extracellular recordings from DSGCs indicated that direction selectivity was mediated primarily by inhibition activated by nulldirection image motion. Strong support for the inhibitory model was provided by subsequent pharmacological experiments, which showed that GABA A -receptor antagonists abolish direction selectivity (Wyatt and Daw, 1976;Ariel and Daw, 1982;Kittila and Massey, 1997). However, these extracellular recording experiments provided no information about whether the inhibition acted directly on the DSGC or presynaptically on the excitatory interneurons.Torre and Poggio (1978) proposed a postsynaptic model in which DSGCs receive an inhibitory input that is spatially offset relative to the excitatory input; moreover, the inhibition is nondirectional, being activated equally well by image motion in the preferred and null directions. During null-direction motion, the spatial offset means that delayed inhibitory synapses at locations ahead of the stimulus are activated and veto the excitation as the stimulus sweeps across the receptive field. For preferreddirection motion, the inhibition trails behind the stimulus and thus arrives too late to veto the excitation. For this model to work, the inhibition must act locally within the dendritic arbor of the DSGC. This occurs when the inhibitory reversal potential is at, or close to, the resting potential of the cell, and therefore the inhibitory input does not polarize the c...