Phylogenetic analyses of toxin gene families have revolutionised our understanding of the origin and evolution of reptile venoms, leading to the current hypothesis that venom evolved once in squamate reptiles. However, because of a lack of homologous squamate non-toxin sequences, these conclusions rely on the implicit assumption that recruitments of protein families into venom are both rare and irreversible. Here we use sequences of homologous non-toxin proteins from two snake species to test these assumptions. Phylogenetic and ancestral-state analyses revealed frequent nesting of 'physiological' proteins within venom toxin clades, suggesting early ancestral recruitment into venom followed by reverse recruitment of toxins back to physiological roles. These results provide evidence that protein recruitment into venoms from physiological functions is not a one-way process, but dynamic, with reversal of function and/or co-expression of toxins in different tissues. This requires a major reassessment of our previous understanding of how animal venoms evolve.