Résumé : We present a statistical test for the hypothesis of rational utility maximization on the basis of nonparametric revealed preference conditions. Our test is conservative for the utility maximization hypothesis. We take as null hypothesis that the consumer behaves randomly, and as alternative hypothesis that she is approximately utility maximizing. Our statistical test uses a permutation method to operationalize the principle of random consumer behavior. We show that the test has an asymptotic power of one against the alternative hypothesis of approximately utility maximizing behavior. We also provide simulated power results and two empirical applications (to experimental and observational data, respectively).