We consider the problem of reconciling two unordered databases whose contents are related. Specifically, we wish to determine the mutual difference of these databases with a minimum communication complexity. This type of problem arises naturally in the context of gossip protocols. We analyze two instances of the reconciliation problem, a client-server model and a more general peer-to-peer model, and provide interactive solutions for both. For the former instance, we also provide a simple one-message reconciliation algorithm, based on elementary symmetric polynomials, which has an almost optimal communication complexity.