Although validation studies show that theoretical models for predicting the pKas of ionizable groups in proteins are increasingly accurate, a number of important questions remain: (1) What factors limit the accuracy of current models? (2) How can conformational flexibility of proteins best be accounted for? (3) Will use of solution structures in the calculations, rather than crystal structures, improve the accuracy of the computed pKas? and (4) Why does accurate prediction of protein pKas seem to require that a high dielectric constant be assigned to the protein interior? This paper addresses these and related issues. Among the conclusions are the following: (1) computed pKas averaged over NMR structure sets are more accurate than those based upon single crystal structures; (2) use of atomic parameters optimized to reproduce hydration energies of small molecules improves agreement with experiment when a low protein dielectric constant is assumed; (3) despite use of NMR structures and optimized atomic parameters, pKas computed with a protein dielectric constant of 20 are more accurate than those computed with a low protein dielectric constant; (4) the pKa shifts in ribonuclease A that result from phosphate binding are reproduced reasonably well by calculations; (5) the substantial pKa shifts observed in turkey ovomucoid third domain result largely from interactions among ionized groups; and (6) both experimental data and calculations indicate that proteins tend to lower the pKas of Asp side chains but have little overall effect upon the pKas of other ionizable groups.