A goal of the modern protein chemist is the design of novel proteins with specific activities or functions. One hurdle to overcome is the ability to accurately predict the pKas of ionizable groups upon their burial in the interior of a protein, where they are typically perturbed from their intrinsic pKas. Most discussion of intrinsic pKas is based on model compound data collected prior to the 1960's. We present here a new set of intrinsic pKas based on model peptides, which we think are more applicable than the model compound values. We observe some differences with the model compound values, and discuss these by critically examining the compounds originally used for the dataset. One interaction affecting the pKas of ionizable groups in proteins that is not well understood is the effect of hydrogen bonds. The side chain carboxyl of Asp33 in RNase Sa is buried, forms 3 intramolecular hydrogen bonds, and has a pKa of 2.4 in the folded protein. One of these hydrogen bonds is to the side chain hydroxyl of Thr56. We mutated Thr56 to alanine and valine and observed that the mutations relieves the perturbation on the carboxyl group and elevates its pKa by 1.5 and 2 units, respectively. The side chain carboxyl of Asp76 in RNase T1 is completely buried, forms 3 intramolecular hydrogen bonds to other side chain groups, and has a pKa of 0.5 in the folded protein. Mutating any of the hydrogen bonding groups to the carboxyl affects its pKa differently, depending on the group mutated. Mutating all of the hydrogen bonding groups, creating a triple mutant of RNase T1, reverses the perturbation on the pKa and elevates it to about 6.4, very near the observed pKa of other carboxyl groups buried in hydrophobic environments. We compared these experimental results with predicted results from theoretical models based on the Solvent Accessibility Corrected Tanford- Kirkwood Equation and the finite difference solution to the linearized Poisson- Boltzmann Equation. The comparisons revealed that these models, most often used by theoreticians, are flawed when typically applied, and some possible improvements are proposed.