Classification task is a popular data mining task in various fields – health, finance, biology etc. and many methods have been developed to solve it. One of the most popular approaches is using decision trees for classification because building models does not require much resources and the models are interpretable for experts who do not know these methods. The article examines possibilities of using meta-information about data structure in data mining classification task. Meta-information can be used in choosing the most suitable classifier and its parameters as well as building a classifier that fits the data. The research also outlines a model of a classifier selection system that is based on the use of meta-information about data. The research analyzes two methods that use information about data structure in building classifiers to design classifiers that fit data, are compact and accurate. Class decomposition uses hierarchical classification to describe class structures by dividing them into subclasses to increase efficiency of classifiers. Attribute-value taxonomy based decision tree design uses structure of attribute values to build classifiers that use attribute values in different abstraction levels. The impact of the use of data structure meta-information on classifiers is experimentally proven using real-life data. The use of both methods significantly improves the performance of classifiers when compared to classifiers that were built without the use of these methods. This proves the utility of meta-information. The article also examines complications than can arise when building meta-information based classifier selection systems and gives recommendations for future work.