Ab initio and other computational studies of bacterial reaction center cofactors are usually performed using the observed (low-resolution) X-ray structures. Unfortunately, these geometries are necessarily approximate and this can have dramatic influences on calculated properties. For example, the calculated energies of the four bacteriochlorophylls in Rhodobacter sphaeroides vary by over 160 kcal mol(-1). To overcome this problem, a combined quantum mechanical/molecular mechanical (QM/MM) method is employed to optimize the structure of the special pair and other cofactors in the photosynthetic reaction centers of Rhodopseudomonas viridis and Rhodobacter sphaeroides, while a purely MM model is used to refine the structure of the remaining protein environment. Specifically, the QM/MM optimizations are performed using a semiempirical AM1-based formalism. After relaxation, the energies of the bacteriochlorophylls differ by only typical conformer relative energies, ca. 5 kcal mol(-1). Another example of improved cofactor properties is the P-L-P-M interaction energy which has been predicted to be strongly repulsive at the X-ray structure but here is shown to be realistically attractive after optimization. After optimization, the distortions in the geometries of the cofactor are seen to be controlled by protein-cofactor interactions, and the cofactors on the L-side are all seen to fit more snugly together within the protein environment than do their M-side counterparts. Also, the 2a-acetyl group of P-M for Rb. sphaeroides, for which hydrogen bonding to the protein is restricted, is predicted to form a weakly bound sixth ligand to the magnesium of P-L; this is consistent with, but not obvious from, the X-ray structure.