A popular view is that the brain works in a similar way to a digital computer or a Universal Turing Machine by processing symbols. Psychophysical experiments and our amazing capability to recognize complex objects (like faces) in different light and context conditions argue against symbolic representation and suggest that concept representation related to similarities may be a more appropriate model of brain function. In present work, by looking into anatomical and neurophysiological basis of how we classify objects shapes, we propose to describe computational properties of the brain by rough set theory (Pawlak, 1992 [1]). Concepts representing objects physical properties in variable environment are weak (not precise), but psychophysical space shows precise object categorizations. We estimate brain expertise in classifications of the object’s components by analyzing single cell responses in the area responsible for simple shape recognition ([2]). Our model is based on the receptive field properties of neurons in different visual areas: thalamus, V1 and V4 and on feedforward (FF) and feedback (FB) interactions between them. The FF pathways combine properties extracted in each area into a vast number of hypothetical objects by using “driver logical rules”, in contrast to “modulator logical rules” of the FB pathways. The FB pathways function may help to change weak concepts of objects physical properties into their crisp classification in psychophysical space.