THE adaptive significance of the widespread protein polymorphism found in natural populations continues to be a central issue in population genetics1–3. Whether this variation is maintained by some form of balancing selection (the selectionist hypothesis) or by the stochastic interaction between selectively neutral mutations and genetic drift (the neutralist hypothesis) remains unclear. Also, evidence is accumulating that the simple electrophoretic separation of proteins detects a much smaller amount of the actual genetic variation present than had been assumed4–9. This has serious implications for much of the data used as evidence in the selection–neutrality controversy. If electrophoretic allozymes are actually large heterogeneous classes of alleles (that is, electromorphs10), then much of the most convincing data presented as evidence for the selectionist hypothesis may be illusory9. Therefore, methods are needed to distinguish between the selectionist and neutralist models without being dependent on detecting all or most of the genetic variation at a particular locus. I describe here a method meeting this criterion by using the relative rate of loss of duplicate gene expression to test the predictions of these two conflicting hypotheses. I report that the rate of loss for different proteins following tetraploidy in two independent groups of fish is positively correlated with the tendency for particular proteins to be polymorphic in natural populations of animals. This relationship is predicted by the ‘neutralist’ hypothesis of protein polymorphism and is in direct conflict with the predictions of the ‘selectionist’ hypothesis.