Abstract
Cognitive activity of a human being is strongly related to the level of symbolic notions rather than numerical values. Bearing this in mind we discuss the use of fuzzy (linguistic) levels in construction of a cognitive structure for pattern recognition. Two structures of neural networks are discussed namely direct and reference scheme of transformation is considered. It is explained how fuzziness is incorporated in the structure of the net and its influence on matching process is clarified. Also interesting aspects of matching and inverse matching procedures are underlined. A question of precision of classification results produced by the network is investigated in detail. Finally numerical results are included.