Bacterial flagellar motility, an important virulence factor, is energized by a rotary motor localized within the flagellar basal body. The rotor module consists of a large framework (C-ring), composed of the FliG, FliM and FliN proteins. FliN and FliM contacts the FliG torque ring to control the direction of flagellar rotation. We report that structure-based models constrained only by residue coevolution can recover the binding interface of atomic X-ray dimer complexes with remarkable accuracy (ca. 1 Å RMSD). We propose a model for FliM-FliN heterodimerization, which agrees accurately with homologous interfaces as well as in-situ cross-linking experiments, and hence supports a proposed architecture for the lower portion of the C-ring. Furthermore, this approach allowed the identification of two discrete and interchangeable homodimerization interfaces between FliM middle domains that agree with experimental measurements and might be associated with C-ring directional switching dynamics triggered upon binding of CheY signal protein. Our findings provide structural details of complex formation at the C-ring that have been difficult to obtain with previous methodologies and clarify the architectural principle that underpins the ultra-sensitive allostery exhibited by this ring assembly that controls the clockwise (CW) or counterclockwise (CCW) rotation of flagella.