This paper critically examines Moscovici's theory of social representations. Four related difficulties are identified and illustrated drawing upon studies by Herzlich (1973), Di Giacomo (1980) and Hewstone et al. (1982). First, it is argued that the treatment of representations as coextensive with social groups is potentially circular and, second, that the broad notion of consensus across representations fails to distinguish important differences between levels (‘use/mention’, ‘theory/practice’). In addition it is claimed that not enough attention is paid to the relationship between social representations and specific contexts of use, and to the constitution of social representations in language. In response to these problems it is suggested that social representations be considered as ‘linguistic repertoires’.