The original feedforward model of Hubel and Wiesel (1962) implicitly consists of two processing stages. The first, the linear summation of input from relay cells whose receptive fields are arranged in rows, has been addressed by many of the experiments discussed so far. The second stage is the nonlinear filtering of the summed inputs by the spike threshold. Threshold is critical to the model in its simplest form: Even when a stimulus is at right angles to the preferred orientation, it will activate a few geniculate neurons since they themselves are not orientation selective. Hubel and Wiesel therefore invoked threshold to prevent a simple cell from responding to these low amplitude inputs (though other mechanisms, such as inhibition, could also contribute, as discussed below). Until recently, the experiment that could directly test the effects of threshold on orientation selectivity, that is, a comparison of the orientation tuning of the membrane potential with the orientation tuning of spike responses in the same neurons, had not been performed. Doing so requires intracellular recordings that are very stable, and, more importantly, that perturb the relationship between spike frequency and membrane potential as little as possible. The advent of in vivo patch recording (Jagadeesh et al., 1992, Pei et al., 1991) has made such recordings possible, even for the small neurons of layer 4. In a recent experiment, Carandini and Ferster (2000) have shown that the orientation tuning of the spike responses is indeed significantly narrowed relative to the tuning of the synaptic inputs. These authors found that the average half-width at half height of the orientation tuning curve for membrane potential in simple cells was 65% greater than that of the spikes (38 vs. 23), suggesting that threshold plays a significant role in shaping the responses of the simple cells.