We study a seller who does not know the distribution of buyers’ values, but can learn from observing their bids in an auction. Which auction formats provide better information about the value distribution? We show that among a large class of standard auctions (e.g., kth-price, all-pay), the first-price auction is (Lehmann) most informative. Thus, while all these auction formats yield the same expected static revenue, the first-price auction is preferred by a seller who can use today’s bid observations to optimize revenue tomorrow.