The feedback models can account for many experimental observations. 1) The most prominent observation is the contrast invariance of orientation selectivity (Skottun et al., 1987, Sclar and Freeman, 1982). 2) Feedback models explain an experiment in which it was found that the orientation tuning of synaptic potentials in some cortical cells sharpened over time (Pei et al., 1994), presumably as the cortical circuitry took over and sharpened the input from the LGN. Note, however, that the feedforward model with push-pull inhibition also predicts that LGN synaptic potentials are sharpened by inhibition at high contrasts. Other time-dependent changes in orientation preferences have been seen extracellularly in monkey visual cortex, though these effects may in part involve input from beyond the classical receptive field (Ringach et al., 1997). 3) The feedback models are roughly consistent with the observed orientation tuning of excitatory and inhibitory inputs to simple cells (Anderson et al., 1999, Ferster, 1986, Nelson et al., 1994, Douglas et al., 1991). Strong inhibition at the preferred orientation is required to keep the excitatory feedback under control, while only weak inhibition at the orthogonal orientation is required to counteract weak excitation. However, the feedback models predict that the inhibition received by a cell should have broader orientation tuning than the excitation it receives. Experimentally, excitation and inhibition appear to have very similar orientation tuning, although small differences cannot be ruled out. 4) Local inactivation of the cortical circuit with injections of GABA can disrupt orientation selectivity of cells hundreds of microns away (Eysel et al., 1990, Crook et al., 1995). This behavior is easily understood when it is the local cortical circuitry that determines orientation selectivity, but is less easily understood in the context of the feedforward models, where the thalamic input is dominant. 5) Because intracortical inhibitory connections are critical to specifying orientation selectivity in the feedback models, experiments in which GABA-antagonists disrupt orientation selectivity are easily understood (Pfluger and Bonds, 1995, Tsumoto et al., 1979, Sillito et al., 1980, Sillito, 1975, Bonds, 1989). 6) The natural tendency of the cortex to amplify small inputs and converge to one of the stable attractors (the stereotyped patterns of activity) brings to mind experiments of Arieli et al. (1995), who found that even in the absence of a stimulus, waves of activity appear and propagate across the cortical surface.