As the improvement in electronics and communication technology, unmanned aerial vehicles (UAVs) play a prominent part for task executions in numerous scenarios. Spectrum resource is seen as a significant factor which may affect the cooperation of UAVs. In this article, the spectrum allocation for task-driven UAV communication networks is investigated. Due to the task-driven essence, different kinds of task lead to different coupling relationships between UAVs in spectrum allocation. By exploring the task-driven essence, some featured challenges and requirements for task-driv-en spectrum allocation are discussed. Following the attractive fact that game theory can model the complex interactions among multiple players, the game-based optimization framework, which consists of task decomposition, task-driven feature requirements, game formulation and distributed algorithm, is developed to tackle the task-driven spectrum allocation problem. Next, some featured game models are proposed and some case studies are presented and discussed. Finally, future discussion and open issues are given to shed light on future investigation.