Descriptivists' method of naturalizing moral language is neither the only nor the most promising one in metaethics. The paper deals with attempts to combine the expressivistic account of moral concepts with an evolutionary research programme. As Allan Gibbard (1990, p. 70) puts it: 'Normative discussion is part of nature but it does not describe nature'. First, Gibbard's expressivism is outlined against the background of the theory of evolution. Then the Author proceeds to his own metaethical theory according to which, to take but one example, the judgment that 'a' is morally wrong consists of a belief that it is possible to avoid a, a belief that there is a universal property 'P' which 'a' exemplifies, a desire not to actualize 'a', a disposition (1) to desire not to actualize anything that instantiates the property 'P', and a disposition (2) to desire to subject everyone who does 'a' to coercive measures (including punishment). Language thus interpreted is shown to be an opposite tool for negotiating a stable normative consensus; it addresses specific problems of cooperation viewed from the evolutionary perspective. Finally, it is argued that the Author's proposal exhibits some important advantages over Gibbard's theory.