Understanding how the variability of protein structure arises during evolution and leads to new structure-function relationships ultimately promoting evolutionary novelties is a major goal of molecular evolution and is critical for interpreting genome sequences. We addressed this issue using the ecdysone receptor (ECR), a major developmental factor that controls development and reproduction of arthropods. The functional ECR is a heterodimer of two nuclear receptors: ECR, which binds ecdysteroids, and its obligatory partner ultraspirade (USP), which is orthologous to the retinoid X receptor of vertebrates. Both genes underwent a dramatic increase of evolutionary rate in Mecopterida, the major insect terminal group containing Dipteras and Lepidopteras. We therefore questioned the implication of this event in terms of coevolution of their dimerization interface. A structural comparison revealed a 30% larger ligand-binding domain (LBD) heterodimerization surface in the Lepidoptera Heliothis when compared with basal insects, associated with a symmetrization of the interface, which is exceptional for nuclear receptors. Reconstruction of ancestral sequences and homology modeling of the ancestral Mecopterida ECR-USP reveal that this enlarged dimerization surface is a synapomorphy for Mecopterida. Furthermore, we show that the residues implicated in the new dimerization surface underwent specific evolutionary constraints in Mecopterida indicative of their new and conserved role in the dimerization interface. Most of all, the novel surface originates from a 15 degrees torsion of a subdomain of USP LBD toward its partner ECR, which is a long-range consequence of the peculiar position of a Mecopterida-specific insertion in loop L1-3, located outside of the interaction surface, in a less crucial domain of the partner protein. These results indicate that the coevolution between ECR and USP occurred through a novel mechanism of intramolecular epistasis that will undoubtedly be generalized for other molecules because it uses flexibility of a less-constrained region of a protein to modify the structure of another, critical part of the molecule.