Realists, D. M. Armstrong among them, claim, contrary to natural class nominalists, that natural classes are analysable. Natural classes of particulars, claim the realists, can be analysed in terms of particulars having universals in common. But for the realist, there are also natural classes of universals. And if the realist's claim that natural classes are analysable is a general claim about natural classes, then the realist must also provide an analysis of natural classes of universals. For Armstrong, the unity (or (...) naturalness) of a natural class of universals is analysed in terms of universals resembling each other. I argue that Armstrong's account fails. His account fails for the same reason all other resemblance accounts of natural classes fail: some arbitrary classes satisfy the analysis for natural classes. (shrink)