We consider a model of state-contingent goods allocation. Each agent has a set of beliefs, or priors, and evaluates an allocation according to the minimum expected utility over the set of priors. We study allocation rules that satisfy efficiency, individual rationality, and strategy-proofness (no one can benefit by misrepresenting his preferences). In the aggregate certainty case, the three axioms are compatible and we characterize the non-empty family of rules which contains both Walrasian and non-Walrasian rules. In the aggregate uncertainty case, we show in the 2-agent and 2-state case that the three axioms may or may not be compatible, depending on existence of a degenerate prior.