Proteins within a molecular network are expected to be subject to different selective pressures depending on their relative hierarchical positions. However, it is not obvious what genes within a network should be more likely to evolve under positive selection. On the one hand, only mutations at genes with a relatively high degree of control over adaptive phenotypes (such as those encoding highly connected proteins) are expected to be “seen” by natural selection. On the other hand, a high degree of pleiotropy at these genes is expected to hinder adaptation. Previous analyses of the human protein–protein interaction network have shown that genes under long-term, recurrent positive selection (as inferred from interspecific comparisons) tend to act at the periphery of the network. It is unknown, however, whether these trends apply to other organisms. Here, we show that long-term positive selection has preferentially targeted the periphery of the yeast interactome. However, the opposite trend is found in flies: genes under positive selection encode significantly more connected and central proteins in the fly interactome. These observations are not due to covariation of genes’ adaptability and centrality with protein length, protein abundance, expression level or, in the case of Drosophila, expression breadth. These results indicate that the distribution of proteins encoded by genes under recurrent positive selection across protein–protein interaction networks varies from one species to another.