The cortex is modeled as a net of minicolumns. A cell assembly in layer IV of a minicolumn is defined as a “cognit”. Higher level cognits integrate lower level cognits. The minicolumn is able to form cognits and extract determinant cognit time series as perceptual symbols. A frame is a set of determinant cognit time series, and simulation is the partially retrieval of these determinant cognit time series. This model can fulfill the requirements for the perceptual symbol system proposed by Barsalou in 1999. I am trying to build neural network simulating this net. It is potentially a Cognitive architecture bearing connectionism.