Protein-protein interactions in solution are often dominated by electrostatics, yet the underlying protein charge is remarkably hard to pin down. Direct experimental methods require complex setups and report only a model-dependent effective charge, so computational prediction methods are commonly used instead - yet these frequently disagree with one another, making their reliability difficult to assess. Here we show that the effective protein-protein pair interaction potential (PIP), measured in the native solution environment by cryogenic electron tomography and predicted from an amino-acid-level coarse-grained model, provides a quantitative benchmark for these methods. Across three globular proteins and a range of pH and buffer conditions, we find that
the orientationally averaged PIP is, to within our resolution, principally governed by the net charge. The predicted net charge is thus the dominant determinant of the modeled PIP, and comparison with the measured PIP makes it possible to assess the schemes that assign charges to the protein. The comparison is sensitive enough to detect subtle, buffer-specific shifts, which we independently trace to direct buffer binding by NMR. Our approach turns the measured PIP into direct experimental feedback on charge prediction and offers a route to refining coarse-grained models of protein interactions.