Abstract
According to a simple similarity theory of representation, x represents y because x and y share some properties. In Meaning and Mental Representation, Robert Cummins rejects this account for representations that play a role in cognition because, among other things, a similarity theory of representation precludes a satisfactory account of an essential cognitive task, namely abstraction. Intelligent beings have representations of classes and properties, and we need an account for such representations. Cummins argues that a causal covariance theory of representation, even a very simple version of it, is superior to a similarity theory because causal covariance at least makes abstraction possible. I want to argue that this is not so. Covariance theories of representation do not achieve abstraction.