The notion of collocation is quite ambiguous. A concise survey of different approaches to it (British contextualism, lexicographical approach, approach of the “Meaning-Text” theory) is proposed in the paper. The paper discusses the results of retrieving collocations from a corpus of Russian texts. The data obtained is compared to the data given for set expressions in modern Russian dictionaries. The paper also explores the role of statistical measures for extracting collocations in Russian, and the issue of their applicability to the Russian language.