A new converse bound is presented for the two-user multiple-access channel under the average probability of error constraint. This bound shows that for most channels of interest, the second-order coding rate-that is, the difference between the best achievable rates and the asymptotic capacity region as a function of blocklength n with fixed probability of error-is O(1/ √ n) bits per channel use. The principal tool behind this converse proof is a new measure of dependence between two random variables called wringing dependence, as it is inspired by Ahlswede's wringing technique. The O(1/ √ n) gap is shown to hold for any channel satisfying certain regularity conditions, which includes all discrete-memoryless channels and the Gaussian multiple-access channel. Exact upper bounds as a function of the probability of error are proved for the coefficient in the O(1/ √ n) term, although for most channels they do not match existing achievable bounds.This result asserts that achievable second-order bounds of [13]-[18] are order-optimal; that is, the gap between the capacity region and the blocklength-n achievable region, in either direction, is at most O(1/ √ n). We provide a specific upper bound on the coefficient in the O(1/ √ n) term, although for most channels it does not match the achievability bounds. The main difficulty in proving a second-order converse for the MAC is to properly deal with the independence between the transmitters. The problem variant with degraded message sets, as studied in [19], [20], seems to be easier precisely because the transmitted signals are not independent. The independence that is inherent to the standard MAC prohibits many of the methods to prove second-order converses for the point-to-point channel; for example, one cannot restrict the inputs to a fixed type (empirical distribution), which is one of the steps in the point-to-point converse in [12], since imposing a fixed joint type on the two input signals creates dependence. An alternative approach adopted in [23] to prove second-order converses uses the notion of reverse hypercontractivity. This technique provides a strengthening of Fano's inequality, wherein the coding rate is upper bounded by the mutual information plus an O(1/ √ n) error term. However, this technique relies on the geometric average error criterion, which is stronger than the usual average error criterion (but weaker than the maximal error criterion). The method of [23] can be applied to the average error criterion by first expurgating the code-i.e., removing some of the codewords with the largest probability of error. However, with the MAC, we cannot just expurgate codewords, we must expurgate codeword pairs, which again introduces some dependence between inputs. For this reason, reverse hypercontractivity can be viewed as a replacement for the blowing-up lemma or Augustin's converse, but does not remove the need for wringing. Interestingly, the technique that we use here seems to be related to hypercontractivity; see Sec. III-D for more details.To handle the ind...