When preschoolers decide to trust one speaker over another, how does group membership influence their tracking of speaker reliability? In Experiment 1, 4-year-olds were assigned to arbitrary groups of no social significance (0055 and 0170) and asked to endorse novel object labels provided by two ingroup members, one of whom was reliable and the second of whom was unreliable. Children selectively trusted the more reliable informant. In Experiment 2, we asked whether ingroup status or reliability would determine children's choices and found that 4-year-olds failed to trust reliable outgroup members over unreliable ingroup members (or vice versa). Experiment 3 showed that the failure of trust in Experiment 2 was not due to the mere inclusion of both ingroup and outgroup members: children presented with a control paradigm in which the ingroup members were reliable trusted reliable ingroup members over unreliable outgroup members. Children's use of reliability as an indicator of future credibility therefore appears disrupted when outgroup status and reliability are in conflict, even when group membership is arbitrary.