ATP, the universal energy currency of cells, is produced by F-type ATP synthases, which are ancient, membrane-bound nanomachines. F-type ATP synthases use the energy of a transmembrane electrochemical gradient to generate ATP by rotary catalysis. Protons moving across the membrane drive a rotor ring composed of 8-15 c-subunits(1). A central stalk transmits the rotation of the c-ring to the catalytic F-1 head, where a series of conformational changes results in ATP synthesis(2). A key unresolved question in this fundamental process is how protons pass through the membrane to drive ATP production. Mitochondrial ATP synthases form V-shaped homodimers in cristae membranes(3). Here we report the structure of a native and active mitochondrial ATP synthase dimer, determined by single-particle electron cryomicroscopy at 6.2 angstrom resolution. Our structure shows four long, horizontal membrane-intrinsic alpha-helices in the a-subunit, arranged in two hairpins at an angle of approximately 70 degrees relative to the c-ring helices. It has been proposed that a strictly conserved membrane-embedded arginine in the a-subunit couples proton translocation to c-ring rotation(4). A fit of the conserved carboxy-terminal a-subunit sequence places the conserved arginine next to a proton-binding c-subunit glutamate. The map shows a slanting solvent-accessible channel that extends from the mitochondrial matrix to the conserved arginine. Another hydrophilic cavity on the lumenal membrane surface defines a direct route for the protons to an essential histidine-glutamate pair(5). Our results provide unique new insights into the structure and function of rotary ATP synthases and explain how ATP production is coupled to proton translocation.