In an innovation contest, an organizer announces an innovation-related problem to a group of agents along with a set of contest rules, and agents develop their solutions (as individuals or as teams) and submit them to the organizer to receive an award. In this thesis, we study innovation contests on crowdsourcing platforms using game-theoretic models in order to generate insights into how organizers should design these contests. In the first paper, we study the optimal contest duration and the optimal award scheme. We show that the optimal contest duration and the optimal total award increase with the agent’s output uncertainty and decrease with the agent’s effort coefficient, which is consistent with recent empirical findings. More interestingly, we show that it is optimal for the organizer to give multiple awards when the organizer has low urgency in obtaining solutions, which helps explain why many contests on platforms give multiple awards. In the second paper, we study team collaboration in innovation contests. We show that agents can benefit from team collaboration when team members are less likely to conform each other and when the organizer’s problem requires some indivisible tasks. We also show that the organizer may benefit from team collaboration when he aims to obtain high-novelty solutions or low-novelty solutions that require mostly divisible tasks, which may explain the mixed policies on platforms regarding team collaboration. In the third paper, we plan to study the impacts of participation rules in crowdsourcing platforms.