The theory of epitaxial strain energy is extended beyond the harmonic approximation to account for large film/substrate lattice mismatch. We find that for fcc noble metals (i) directions 001 and 111 soften under tensile biaxial strain (unlike zincblende semiconductors) while (ii) 110 and 201 soften under compressive biaxial strain. Consequently, (iii) upon sufficient compression 201 becomes the softest direction (lowest elastic energy), but (iv) 110 is the hardest direction for large tensile strain. (v) The dramatic softening of 001 in fcc noble metals upon biaxial tensile strain is caused by small fcc/bcc energy differences for these materials. These results can be used in selecting the substrate orientation for effective epitaxial growth of pure elements and ApBq superlattices, as well as to explain the shapes of coherent precipitates in phase separating alloys. When a material is compressed hydrostatically, its energy rises steeply because all three crystal axes are deformed (dashed line in Fig. 1). When the same material is confined coherently onto a substrate ("coherent epitaxy") with lattice constant a s , the energy rises less steeply (solid line in Fig. 1) since it is deformed only along the crystal axes in the substrate plane and allowed to relax (and thus, lower its energy) in the third direction G. This "epitaxial softening" can be quantified by the dimensionless parametergiving the ratio between the epitaxial increase in energy due to biaxial deformation to a s , and the hydrostatic increase in energy due to triaxial deformation to the same a s . Because the biaxial strain energy ∆E epi (a s , G) depends on the strain direction G, so does q(a s , G). In growing coherent epitaxial films, 1 it is desirable to minimize ∆E epi (a s , G) [or, equivalently, for a fixed a s minimize q(a s , G)], so as to avoid or reduce dislocations and other strain-induced film/substrate defects. It is hence important to select substrates a s and growth directions G that entail minimal strain energy. Harmonic continuum elasticity theory 2,3,4,5,6,7,8,9 makes definitive predictions for the a s -and G-dependence of epitaxial strain energy: (i) q(a s , G) does not depend on a s , and (ii) the "softest direction" is 001 if ∆ = [C 44 − 1 2 (C 11 − C 12 )] > 0, while if ∆ < 0 then the softest direction is 111 . For most fcc metals and semiconductors ∆ > 0, whereas ∆ < 0 for ionic salts (PbS, AgBr, NaCl, KCl) and several bcc metals (Nb, V, Mo, Cr). The selection of substrate orientation for many years has been guided by these predictions of harmonic elasticity, summarized compactly by the expression 6,7