This paper is devoted to investigate efficiency of thesauri use in popular natural language processing (NLP) fields: information retrieval and analysis of texts and subject areas. A thesaurus is a natural language resource that models a subject area and can reflect human expert's knowledge in many NLP tasks. The main target of this survey is to determine how much thesauri affect processing quality and where they can provide better performance. We describe studies that use different types of thesauri, discuss contribution of the thesaurus into achieved results, and propose directions for future research in the thesaurus field.