The task of pattern recognition is to assign an input pattern to one of the known classes of classified patterns. The idea of the syntactic approach to pattern recognition is to decompose the complex input pattern into a hierarchy of simpler subpattems and to develop rules by which they can be combined to assemble a high-level pattern. The decomposing of the original pattern is called parsing and the rules to combine the subpattems are based on formal language theory. This theory works with concepts like words, sentences and syntactic rules which are borrowed from ordinary languages. The input patterns comprise of features. One of the steps within a pattern recognition problem is the feature extraction. The feature selection is very important for pattern recognition : the feature set is selected to ensure a good separation of pattern classes. The features are combined to form patterns according to the pattern grammar rules in a way similar to the human languages. Similarly, some combinations are allowed ,others not. The syntactic approach to the pattern recognition problem works with the following notions: S the input patterns are called words ,which are strings offeatures (same as letters from an alphabet); S there is a grammar for each class ofpatterns , which generates the objects from the class as part ofthe language. . parser is responsible to decide the validity of a pattern , seen as a word from the language generated by the grammar. What seems to be an easy way to classify the input objects based on syntactic rules, is unfortunately not. The instrumentation as well as the algorithm errors and approximations are leading to a much more complicated task. To be able to ensure a good separation of classes , the dimension ofthe feature set has to be increased ; therefore the complexity of the algorithms and the computational time will increase to unacceptable values. The paper presents a solution using the attributed grammars to reduce the feature set dimension. Each feature (seen as a letter from the grammar alphabet) has a set of attributes ; therefore the input pattern , seen as a string of concatenated features ,has a multidimensional structure. Since the grammars used are dealing with multidimensional strings ,the grammars themselves are multidimensional.