"Task-set" is an essential concept, with a long history, but the selection and retrieval of task-sets is poorly understood, and competition between task-sets has rarely been distinguished from competion between individual response tendencies. However, recent research on task-switching costs (or task-repetition benefits) has focused attention on how task-sets are
re-instated.

I will introduce task-switching paradigms with a recent experiment directly comparing predictable and unpredictable switching. Like many others, this study demonstrates the reduction in switch costs observed with an opportunity for preparation (the "preparation effect") and a "residual cost": a component of the switch cost apparently resistant to elimination
by advance preparation. The preparation effect is now generally agreed to index an endogenous task-set reconfiguration process, but has not yet been much exploited to explore its properties; I will describe a sample experiment on the efficacy of different types of cue. The residual cost, because counterintuitive, has excited rather more interest: I will assess interpretations of the residual cost as positive/negative task-set priming, and as the duration of an obligatory or probabilistic post-stimulus control process required on switch trials, attempting to cover data on (a) fitting a mixture model to RT distributions, (b) asymmetries of switch cost, (c) LRPs on switch and non-switch trials, (d) additivities in switch cost, and (e) switch costs with "univalent" stimuli affording only one task.