It is shown that an appearance of an intermediate martensite phase called adaptive martensite may be expected if the surface energy of a boundary between two orientational variants of the normal martensite phase is very low and the typical lattice-mismatch-related elastic energy is high. The adaptive martensite is formed as an elastically constrained phase when the scale of structure heterogeneities induced by the crystal-strain accommodation is reduced to the microscopic scale commensurate with the twin-plane interplanar distance. An example of the cubic\ensuremath{\rightarrow}tetragonal transformation is considered where the adaptive phase has a pseudo-orthorhombic lattice. The crystal-lattice parameters of the adaptive phase are expressed through those of the parent cubic phase and tetragonal normal martensite. It is shown that the (5,2\ifmmode\bar\else\textasciimacron\fi{})7R martensite in \ensuremath{\beta}'NiAl alloys and the intermediate phase recently found just above the temperature of the fcc\ensuremath{\rightarrow}fct martensitic transformation in Fe-Pd are examples of the adaptive martensite. A possible role of the adaptive phase in the thermal nucleation of the martensite is discussed. The nucleation of the normal martensite may be bypassed by nucleation of the adaptive phase, which transforms to the normal martensite during the growth.