The enormous flow of new paintings and their electronic images in the Internet evoke the need of automatization of their primary analysis from the art history point of view, particularly in terms of symbols. The key concept of “symbol” is one of the keystone in the art and one of the most difficult to formalize at the same time. The precedent-based approach seems to be most natural for the formalization. It implies the creation of a training set of images tagged by symbols and considering the tagging problem as a machine learning task. We propose to use the libraries of symbols, signs, and glyphs, arranged by categories such as culture, country, religion, and more, such as http://www.symbolarium.ru in Russian and https://www.symbols.com in English. We propose to use text descriptions of paintings themselves and the articles from the encyclopedia of symbols in order to better associate images and symbols.