Three hypotheses to explain the amino acid composition of proteins are inconsistent (P ≃ 10-9) with the experimental data for β-galactosidase from Escherichia coli. The exceptional length of this protein, 1021 residues, permits rigorous tests of these hypotheses without complication from statistical artifacts. Either this protein is not at compositional equilibrium, which is unlikely from knowledge about other proteins, or the evolution of this protein and its coding gene have not been selectively neutral. However, the composition of approximately 60 percent of the molecule is consistent with either a selectively neutral or nonneutral evolutionary process. Copyright © 1979 AAAS.