Neurophysiological recordings from behaving animals have shown that large numbers of neurons (10^6 or more) have activity patterns that correlate with any given behavior. Current methods that record from individual neurons allow recording from only hundreds to thousands of them, at best. How can we infer how the large circuit works from such a paltry sample of it? One answer is that many of the neurons appear to have responses that are very similar to each other. Dimensionality reduction methods suggest that the effective number of degrees of freedom of the circuit is small enough that we can begin to infer its properties. I will talk about this in the context of a specific example, data taken from the prefrontal cortex of animals during a 3-second period while the animals are still but holding a previously presented stimulus in short-term memory. I will discuss what these results tell us about the neural mechanisms underlying short-term memory.