We develop a principal/agent model for matching agents in two-sided assignments. A principal has preferences over all agents' assignments, and agents have privately-known preferences about their own match (but are indifferent about others'). Unhappy agents can quit, but the principal can stop...