Continuum electrostatic calculations were carried out for a data set of 21 salt bridges using three sets of parameters (CHARMM, PARSE, and a version of of OPLS with slightly exaggerated atomic radii) and three representations of the dielectric boundary (approximate, exact, and smoothed versions of the analytic molecular surface). Comparisons of the overall results and a decomposition into terms due to desolvation, the direct salt bridge, and other interactions with the surrounding protein were used to analyze the effect of differences in parameter sets. It is concluded that (1) while numerical differences among the results can be significant, the underlying tension between desolvation penalty incurred and favorable interactions recovered upon folding is similar in all computations; (2) the largest variation occurs for the desolvation penalty, indicating the importance of fitting this property in parameter development : (3) the approximation of the molecular surface produces the largest desolvation penalties, whereas the smoothed boundary produces the smallest; (4) the exaggerated-radii OPLS implementation used here resulted in underestimates of the desolvation penalty and overestimates of the bridge interaction, relative to the other parameter sets, for 15 of 21 salt bridges, and (5) although the origins and parameter values are different for the CHARMM and PARSE parameter sets, they produce rather similar energetics, due in part to compensating effects.