In this paper, we offer an explanation for how selectivity for orientation could be produced by a model with circuitry that is based on the anatomy of V1 cortex. It is a network model of layer 4C␣ in macaque primary visual cortex (area V1). The model consists of a large number of integrate-and-fire conductance-based point neurons, both excitatory and inhibitory, which represent dynamics in a small patch of 4C␣-1 mm 2 in lateral area-which contains four orientation hypercolumns. The physiological properties and coupling architectures of the model are derived from experimental data for layer 4C␣ of macaque. Convergent feed-forward input from many neurons of the lateral geniculate nucleus sets up an orientation preference, in a pinwheel pattern with an orientation preference singularity in the center of the pattern. Recurrent cortical connections cause the network to sharpen its selectivity. The pattern of local lateral connections is taken as isotropic, with the spatial range of monosynaptic excitation exceeding that of inhibition. The model (i) obtains sharpening, diversity in selectivity, and dynamics of orientation selectivity, each in qualitative agreement with experiment; and (ii) predicts more sharpening near orientation preference singularities.T he mammalian primary visual cortex (area V1) marks the first site along the ''visual pathway'' [Retina 3 Lateral geniculate nucleus (LGN) 3 V1 3 And beyond], where selective response is observed to elementary features of visual scenes, such as orientation and spatial frequency. Despite 40 years of intense research effort, a detailed account of the neural basis for this selectivity in V1 remains elusive. In this paper we focus on orientation selectivity, the selective response of a single neuron to some orientations of a bar or grating, and not to others. This property of single cortical cells was discovered by Hubel and Wiesel (1) in 1962; it is probably important for tasks such as edge detection and contour completion (2). A basic question is still unanswered: to what degree, and by what mechanisms, does cortical processing contribute to orientation selectivity? V1 is a layered structure, with different layers having different tuning properties and functional architectures. Here, we focus on layer 4C␣ because it is an input layer for stimulus from the LGN (magnocellular pathway). Data illustrating examples of orientation selectivity in an input layer in V1, 4C␣, are shown in Fig. 1 (D. Ringach, M. Hawken, and R.S., unpublished work). Fig. 1a shows sample tuning curves for three simple cells in layer 4C␣, in response to a drifting grating oriented at angle (angles separated by 180°designate gratings of the same orientation drifting in opposite directions). These are tuning curves of the steady-state firing rate averaged over many repeated periods of drift. These curves hint at the great diversity observed in the selectivity of 4C␣ neurons. Two neurons show peaks at their ''preferred angles,'' with one weakly and the other more strongly selective, whereas the third ...