Our work focuses on specialized circuitry in the inner retina. Having examined several inner retinal synapses in physiological detail, we now seek to understand how these synapses contribute to visual processing in the surrounding circuitry. We combine electrophysiology, imaging approaches and cellular/network modeling to explore dendritic integration in directionally-selective ganglion cells (DSGCs). Specifically, we have examined how directionally tuned synaptic inhibition, coming from starburst amacrine cells, and NMDA receptor-mediated input from bipolar cells combine in DSGC dendrites to achieve perfectly multiplicative scaling of postsynaptic potentials, enabling directional signals to become larger with no change in directional selectivity. We found that NMDA receptors enable a consistent interaction between excitation and inhibition that enhances the fidelity and robustness of directional signaling (Poleg-Polsky and Diamond, 2016, Neuron). Our work on feedback inhibition has led us to undertake a long-term effort to understand, in a systematic way, how amacrine cells contribute to visual signaling in the inner retina. With over 40 different amacrine cell subtypes, this prospect can be a bit overwhelming. To start, we have examined visual processing in starburst amacrine cells (SACs). In a collaboration with Kevin Briggman (NINDS) and Robert Smith (Penn), we found that SACs in mouse retina receive their synaptic inputs in a different pattern compared with rabbit retina (the classical model for direction selectivity). Numerical simulations predicted that these differences would yield distinctive velocity tuning, something that we confirmed by imaging visually evoked responses in SAC dendrites (Ding, et al., 2016, Nature). We have continued to examine visual signaling in SAC dendrites to identify the fundamental computational unit within the dendritic arbor; a manuscript is in preparation. Robust direction selectivity requires a consistent ratio of excitation to inhibition for a particular direction, regardless of other stimulus parameters (e.g., visual contrast). We found that the excitation:inhibition ratio in DSGCs remains consistent over a range of light intensities. This occurs despite the fact that both DSGCs and SACs receive excitatory input from bipolar cells, and then SACs make feedforward inhibitory inputs onto DSGCs. SACs exhibit a nonlinear input/output relationship, due primarily to the voltage-dependence of voltage-gated calcium channels, that makes SAC output less contrast-sensitive than SAC input. We discovered that the bipolar cell inputs to SACs exhibit higher contrast sensitivity than those to DSGCs, so that the SAC nonlinearity yields an inhibitory signal to DSGC that is contrast-matched to the excitatory input from bipolar cells to DSGCs (Poleg-Polsky and Diamond, 2016, J. Neurosci.).