We present a model of altruistically-minded-yet rational-players contributing to a public good. A key feature is the tension between altruism and "crowding-out" effects (players' efforts are strategic substitutes). We find that more altruistic behaviour can raise or reduce welfare, depending on the fine details of the environment. It is almost always optimal for a player to act more selfishly than her true preference. We discuss applications to a range of public good problems, including global climate policy. Our results highlight that it may be difficult to infer social preferences from observed behaviour.