Abstract. In the standard general-adversary model for multi-party protocols, a global adversary structure is given, and every party must trust in this particular structure. We introduce a more general model, the asymmetric-trust model, wherein every party is allowed to trust in a different, personally customized adversary structure. We have two main contributions. First, we present non-trivial lower and upper bounds for broadcast, verifiable secret sharing, and general multi-party computation in different variations of this new model. The obtained bounds demonstrate that the new model is strictly more powerful than the standard general-adversary model. Second, we propose a framework for expressing and analyzing asymmetric trust in the usual simulation paradigm for defining security of protocols, and in particular show a general composition theorem for protocols with asymmetric trust.