This chapter provides the first empirical test of the claim that people recover meaning from covariation alone. It examines two theories that feature covariation as an important component of learning the meaning of concepts — Launder and Dumais 1997; Rogers et al. 2004 — followed by reasons to question a reliance on covariation alone. It then presents the results of three experiments that demonstrate limits on how much meaning can be recovered from covariation alone. Given these limits, it discusses an issue central to the debate — what kind of data supports the notion that arbitrary, abstract, amodal (AAA) symbols play a role in cognition.