This paper provides a game theoretic model of a crowdsourcing contest. Special attention is given to the asymptotic behavior of the contest outcome. We show that all significant outcomes of crowdsourcing contests will be determined by contestants in a small neighborhood (core) of the most efficient contestant type; in particular, the asymptotic structure of the crowdsourcing contests is distribution-free. Our formal analysis yields a managerially implementable and easily understood rule of thumb for the optimal division of the contest budget among multiple prizes. When agents are risk-neutral, the principal should optimally allocate all of its budget to the top prize even if it values multiple submissions. In contrast, if agents are sufficiently risk-averse, the principal may optimally offer more prizes than the number of submissions it desires. Our paper represents the first general analysis of the economics of crowdsourcing contests, provides a simple rule of thumb for determining the optimal prize structure for practitioners who are considering designing such contests, and also discusses the welfare implications of organizing production or R&D as a Web-based contest of this kind.