An experiment is reported that measured the time it takes to switch attention from one set of locations to another in response to a cue that indicates the relevant locations. The experiment compared sequences of trials in which the same locations were cued in succession with sequences in which different locations were cued in succession in order to separate cue-encoding time from attention-switching time. Same-location sequences require cue encoding but not attention switching. They were substantially faster than different-location sequences, which require both cue encoding and attention switching. Formal models were fitted to time-course functions generated by presenting the cues 0, 100, 200, 300, or 400 msec before the target displays. The model fits suggest that cue encoding took 67-74 msec and attention switching took 76-101 msec.