This paper presents EPAC (Electrostatic isoPotential Analytical Comparative model), the first statistical model for evaluating the geometric similarity of electrostatic fields. Beginning with aligned binding cavities, EPAC measures similarity based on the overlapping volume of isopotentials inside ligand binding cavities. We tested the accuracy of our model on two subfamilies of the serine protease superfamily, demonstrating that EPAC effectively identifies binding sites that prefer differently charged substrates. For example, EPAC identified subtle electrostatic variations in proteins that might be expected to be more similar, such as the difference between typical trypsins and a trypsin with a phosphorylated tyrosine nearby the binding site. These results point to applications in the unsupervised comparison of many binding sites from a purely electrostatic perspective, in the search of subtle electrostatic variations that could influence binding specificity.