What is distributed representation? This question is central to many practical and philosophical concerns, both in connectionism and in cognitive science more generally, yet it has never been given an answer that is both comprehensive and precise. At one time or another researchers have pointed to distribution in a wide variety of places: in the way knowledge is stored in the brain, in connectionist networks, in optical holography, and in psychological models of memory, to name just a few. Is there in fact one general kind of representation encompassing this diversity? Can it be given a reasonably rigorous and useful characterization? This paper sketches answers to these questions. It proposes a way of defining distribution that reveals the key similarity between, for example, the gross representational properties of various brain areas on one hand and connectionist hidden unit activity patterns on the other. Meanwhile, the definition is strict enough to yield mathematically precise descriptions in real modeling contexts.