The crystal structure of mitochondrial malate dehydrogenase from porcine heart contains four identical subunits in the asymmetric unit of a monoclinic cell. Although the molecule functions as a dimer in solution, it exists as a tetramer with 222 point symmetry in the crystal. The crystallographic refinement was facilitated in the early stages by using weak symmetry restraints and molecular dynamics. The R-factor including X-ray data to 1.83-A resolution was 21.1%. The final root mean square deviation from canonical values is 0.015 A for bond lengths and 3.2 degrees for bond angles. The resulting model of the tetramer includes independent coordinates for each of the four subunits allowing an internal check on the accuracy of the model. The crystalline mitochondrial malate dehydrogenase tetramer has been analyzed to determine the surface areas lost at different subunit-subunit interfaces. The results show that the interface with the largest surface area is the same one found in cytosolic malate dehydrogenase. Each of the subunits contains a bound citrate molecule in the active site permitting the elaboration of a model for substrate binding which agrees with that found for the crystalline enzyme from Escherichia coli. The environment of the N-terminal region of the crystallographic model has been studied because the functional protein is produced from a precursor. This precursor form has an additional 24 residues which are involved in mitochondrial targeting and, possibly, translocation. The crystallographic model of mitochondrial malate dehydrogenase has been compared with its cytosolic counterpart from porcine heart and two prokaryotic enzymes. Small but significant differences have been found in the polar versus nonpolar accessible surface areas between the mitochondrial and cytosolic enzymes. Using least squares methods, four different malate dehydrogenases have been superimposed and their consensus structure has been determined. An amino acid sequence alignment based on the crystallographic structures describes all the conserved positions. The consensus active site of these dicarboxylic acid dehydrogenases is derived from the least squares comparison.