Research from different laboratories could be compared more easily if a common set of research tasks were used. Such tasks should be amenable to controlled laboratory research, yet the results should generalize to important real-world tasks. In this paper, we describe the design of synthetic tasks, which are research tasks constructed by systematic abstraction from a corresponding real-world task. We present as an example a laboratory task (the cloud-break task) derived from a particularly demanding part of the reconnaissance mission of the USAF Predator uninhabited air vehicle. We describe potential pitfalls in decoupling a synthetic task from its normal mission context, and discuss some lessons learned from a preliminary design study.