Games of pure mutual interest require players to coordinate their choices without being able to communicate. One way to achieve this is through team-reasoning, asking ‘what should we choose’, rather than just assessing one’s own options from an individual perspective. It has been suggested that team-reasoning is more likely when individuals are encouraged to think of those they are attempting to coordinate with as members of an in-group. In two studies, we examined the effects of group identity, measured by the ‘Inclusion of Other in Self’ (IOS) scale, on performance in nondescript coordination games, where there are several equilibria but no descriptions that a player can use to distinguish any one strategy from the others apart from the payoff from coordinating on it. In an online experiment, our manipulation of group identity did not have the expected effect, but we found a correlation of .18 between IOS and team-reasoning-consistent choosing. Similarly, in self-reported strategies, those who reported trying to pick an option that stood out (making it easier to coordinate on) also reported higher IOS scores than did those who said they tended to choose the option with the largest potential payoff. In a follow-up study in the lab, participants played either with friends or with strangers. Experiment 2 replicated the relationship between IOS and team-reasoning in strangers but not in friends. Instead, friends’ behavior was related to their expectations of what their partners would do. A hierarchical cluster analysis showed that 46.4% of strangers played a team reasoning strategy, compared to 20.6% of friends. We suggest that the strangers who group identify may have been team reasoning but friends may have tried to use their superior knowledge of their partners to try to predict their strategy.