Due to the lack of computational power to perform a fully atomistic simulation of practical, engineering systems, a number of concurrent multiscale methods is developed to limit atomic model to a small cluster of atoms near the hot spot. In this paper the overview of salient features of the main multiscale families is given. The special attention is drawn towards the role of model adaptivity, that is, which part of the problem domain to model by the atomic scale (the hot spot) and which by coarse scale model, as well as where to place the interface of the two models to control the accuracy. Taking Quasicontinuum method as a reference, review of the evolution of the Bridging domain/Arlequin method is given, which parallels the development of a posteriori modeling error estimation.