It has frequently been proposed that stereotypes are self-maintaining at least in part because people tend to better remember expectancy-confirming (versus expectancy-disconfirming) information about social groups. This memory bias is assumed to occur because stereotype-consistent behaviours and traits are more easily associated with the social group label in memory, and thus are more readily activated from memory when thinking about the group or about group members. The results of 26 experiments that studied memory for information describing members of existing social groups were meta-analytically investigated to assess the validity of this hypothesis. As predicted, this analysis revealed an overall consistency effect for both free recall memory and for recognition memory measures that were not controlled for guessing. Analysis of relevant moderating variables suggested that these effects were due to more strongly developed mental associations between expectancy-consistent (versus -inconsistent) information and the group label. Results are discussed in terms of their implications for stereotype maintenance, and for the process of stereotyping.