Though the question is ontological, I will approach it through another, partially linguistic, question. What must natural kinds be like, if the conventional wisdom about natural kind terms is correct? Although answering this question won’t tell us everything we want to know, it will, I think, be useful in narrowing the range of feasible ontological alternatives. I will therefore summarize what I take to be the contemporary linguistic wisdom, and then test different proposals about kinds against it. As we will see, some fare better than others. Following Kripke, I take natural kind terms to be akin to proper names.1 Like names, they are not synonymous with descriptions associated with them by speakers. They are also like names in the way they are introduced, and have their reference fixed. Just as ordinary names are often introduced by stipulating that they are to refer to certain individuals with which one is already acquainted, so general terms are often introduced with the intention that they are to designate natural kinds with which one is acquainted through their instances. For example, we may imagine the terms ‘green’, ‘gold’, ‘water’, and ‘tiger’ being introduced by the following stipulations: The term ‘green’ is to designate the color of (nearly) all members of a certain class of paradigmatic samples – i.e. it is to designate the characteristic of object surfaces causally responsible for the fact that those samples appear similar to us (and different from paradigmatic non-green samples). Hence, the predicate ‘is green’ will apply (at a world-state) to precisely those objects the surfaces of which have the characteristic which, at the actual world-state, causally explains why the green-samples look similar to us.