In a science with very few principles and laws, psychology has offered the principle of least effort: Given a choice between similarly rewarding options, organisms avoid options that require more work or effort. Simply put, effort is costly, and organisms will devalue rewards because of it. Here, I ask whether a basic facet of social cognition—namely empathically feeling what others feel—is effortful and whether people generally avoid it (all else being equal). We develop a new measure of empathy choice called the Empathy Selection Task, where participants make a series of binary choices between situations that involve active empathy or alternative courses of action. The task allows participants to make free choices about which kinds of situations—empathic or non-empathic—they prefer. Across many studies (N = 2,374) we find strong evidence for empathy avoidance, which is associated with perceiving empathy as effortful, aversive, and inefficacious. Further, people avoid empathy for negative and positive states, and this effect is not reducible to emotion avoidance more generally. Empathy avoidance is moderated by the identity of the empathy target (i.e., stranger vs. loved one), consistent with the notion that people will empathize when sufficiently incentivized. These results qualify strong claims that empathy is a default and suggest that when given the choice to feel empathy, people often prefer not to.