Current psychophysical models assume that filter outputs are combined to result in threshold signals. This is sufficient for pattern discrimination but fails to explain how things look. A possible solution of the problem is to use pattern recognition methodologies for analyzing spatial vision. However, internal visual representations may differ from those derived from the dimensions of the physical signal variance. Such deviations can be analyzed by using the principle of "virtual prototypes," i.e., a technique of least-mean-squares adaptive filters.