We use administrative panel data to decompose worker performance into components relating to general talent, task-specific talent, general experience, and task-specific experience. We consider the context of high school teachers, in which tasks consist of teaching particular subjects in particular tracks. Using the timing of changes in the subjects and levels to which teachers are assigned to provide identifying variation, we show that much of the productivity gains to teacher experience estimated in the literature are actually subject-specific. By contrast, very little of the variation in the permanent component of productivity among teachers is subject-specific or level-specific. Counterfactual simulations suggest that maximizing the value of task-specific experience could produce nearly costless efficiency gains on the order of .02 test score standard deviations.