The free energies of radical-pair states of photosynthetic bacterial reaction centers are examined by calculations based on the crystal structure of the reaction center from Rhodopseudomonas viridis. The calculations focus on the energies of the states P+B-H and P+BH-, where P is a bacteriochlorophyll dimer that serves as an electron donor in the photochemical electron-transfer reaction, H is the bacteriopheophytin that accepts the electron, and B is a bacteriochlorophyll that may act as an intermediary. Dielectric effects are treated microscopically by evaluating the induced dipoles on the protein atoms and on a grid of points representing the surrounding membrane and solvent. Calculations using both the crystallographic coordinates for the protein atoms and molecular-dynamics/free-energy-perturbation simulations are carried out with various treatments of the ionizable amino acid residues and with several different models of the membrane. Effects of electrolytes in the solvent are included. The dependence of the results on the size of the protein region that is treated explicitly in the model is examined. Calculations that do not include the membrane or solvent are shown to give unstable results that cannot be used to draw conclusions about the energies of the radical-pair states. On the other hand, accounting properly for the dielectric effects of the protein, membrane, and solvent makes the calculated free energies relatively insensitive to the size of the protein model, the charges assigned to the ionizable amino acid residues, and other details of the treatment. The calculations place P+BH- 6-7 kcal/mol below the excited singlet state of P, in good agreement with experimental measurements, and put P+B-H about 3 kcal/mol above P+BH- with an uncertainty of several kilocalories per mole. These results are consistent with the formation of P+B-H as an intermediate in the charge-separation reaction, although we cannot exclude the possibility that the reaction proceeds by a superexchange mechanism. Most of the ionized amino acid residues probably are sufficiently well screened so that they have only minor electrostatic effects on the energies of the relaxed P+B-H and P+BH- states, but the effects of two arginines and an aspartic acid residue could be significant. Fields from other ionized groups could be important on time scales that are short relative to relaxation of the protein and solvent dipoles. If the solvent is assigned a low polarity in order to model a long dielectric relaxation time, the calculated reorganization energies of the electron-transfer reactions are decreased but our conclusions about the energetics of the radical-pair states are not changed significantly.