We consider a model in which an agent takes actions to affect her reputation with two heterogeneous audiences with diverse preferences. This contrasts with standard models of reputation, which treat the audience as homogeneous. A new aspect that arises with heterogeneous audiences is that different audiences may observe outcomes commonly or separately. We show that, if all audiences commonly observe outcomes, reputation concerns are necessarily efficient-the agent's per-period payoff in the long-run is higher than in one-shot play. However, when the audiences separately observe different outcomes, the result is opposite-the agent's per-period payoff is lower than in one-shot play. Therefore, whenever possible, the agent would choose to deal with audiences commonly rather than separately. If this is not possible, the second-best solution may be to simply forego one's reputation with one of the audiences and focus entirely on the other. Our framework thus lends support to the perception that it is hard to please different diverse audiences, and it may be better to not even attempt to do so.