The preceding analysis has been developed under the assumption of linearity of simple cell responses. Specifically, linearity occurs at two levels:
At behavioral level, indeed, several experimental data on frequency-doubled responses at high spatial frequencies , response saturation at high contrasts [1,113],  non-specific stimulus suppression , proved the existence of simple cells characterized by marked nonlinearities. Such behavioral observations cannot be predicted by our linear model without incorporating threshold nonlinearities  and/or other normalization stages [17,54]. At architectural level, because of the complex ``chemical machine'' of intracortical synaptic microcircuits, linearity is more the exception than the rule. In particular, concerning inhibition, though recent investigations [31,35,36,38] revealed that in striate cortex the predominant mechanism is membrane hyperpolarization (i.e., subtractive action), other mechanisms such as shunting inhibition (i.e., divisive action) or pre-synaptic suppression, are also present [15,66], and they have been indicated as the most probable source of nonlinearities in simple cell responses [10,86]. All these lines of evidence indicate limits to the linearity of the response of simple cells and thus potential errors in our linear analysis. However, considering that only a small percentage of simple cells show a marked nonlinear behavior, and that such behavior occurs only for extreme conditions of input stimulation, to a large extent the linear assumption is a good approximation of reality.