Please cite the published version.ignore his alcoholic friends chooses rationally; we might be a little more circumspect about the rationality of choosing to drink warm beer over drinking something else; and, we might dismiss the decision to serve guests a mixture of red and white wine as absurd. Whatever we think of the rationality of the conclusions, we will surely observe that, in each of the different scenarios, the conclusion is justified by appealing to the claim that the chosen scenario approximates the ideal scenario better than any other alternative. Furthermore, whatever we think about the rationality of the conclusions, we might think that each of the conclusions has been poorly justified. More specifically, we might think that the alternatives that most resemble one another in their "features" are not necessarily, or even likely to be, closest in their desirability.This idea-that closeness in features does not entail closeness in desirability-is often appealed to in moral and political philosophy. 2 Many call it "the problem of second best", but this wrongly suggests that it only applies to situations where one is interested in the second best option. To avoid this implication, we will follow Estlund (2019) and call it "the fallacy of approximation". This article develops a general account of the fallacy of approximation that improves on accounts currently available in the literature. In §1 we explore how many philosophical discussions of the fallacy of approximation appeal to the famous theorem of Lipsey and Lancaster. We argue that these appeals typically misuse the theorem and that a full account of the fallacy of approximation must be developed independently of Lipsey and Lancaster's theorem. In §2 we explore David Estlund's account of the fallacy of approximation-the only systematic account developed in the literature that moves beyond the Lipsey and Lancaster theorem. We argue that, although Estlund's account is an improvement on previous accounts of the fallacy in the literature, it has several serious flaws that make it untenable. Finally, in §3, we develop our own account of the fallacy of approximation which is based on the notion of "relevant descriptive similarity". By capturing the wide variety of contexts in which approximation reasoning is used, our account avoids the problems that undermine Estlund's account. It also shows that there is no simple "fallacy of approximation" but rather several different ways in which approximation reasoning can go wrong. Finally, it shows how reasoning by approximation is a legitimate form of reasoning in many circumstances. Given this, our account makes progress possible in philosophical debates that use approximation reasoning.