Résumé : We provide a revealed preference characterization of equilibrium behavior in first price sealed bid auctions. This defines testable conditions for equilibrium play that are intrinsically nonparametric, meaning that they do not require a non-verifiable specification of the individual utility functions. We characterize equilibrium play for a sequence of observations on private values and bids for a given individual. In a first step, we assume that the distribution of bids in the population is fully known. In a second step, we relax this assumption and consider the more realistic case that the empirical analyst can only use a finite number of i.i.d. observations drawn from the population distribution. We demonstrate the empirical usefulness ofour conditions through an illustrative application to an existing experimental data set of Neugebauer and Perote 2008. This application also shows the potential of our nonparametric characterization to study the behavioral phenomena learning and fatigue at the individual level.