Abstract: Atomic structure is the basis for deep understanding of various physical and chemical properties of condensed matter. In recent years, with the rapid development of global optimization algorithms and first-principles methods, it has become possible to determine or predict the atomic structure through theoretical calculations with only information about the chemical composition and external conditions. This article will briefly introduce the basic principles of current theoretical structure prediction methods, with particular emphasis on the swarm-intelligence based CALYPSO method and its applications.