Refreshing – briefly attending to an item in working memory – has been proposed as a domain-general maintenance process. According the time-based resource-sharing (TBRS) theory, people refresh the contents of working memory sequentially at high speed. We measured the speed of refreshing by asking participants to sequentially refresh a small set of items in sync with a metronome, and to adjust the metronome to the fastest speed at which they could refresh. Refreshing speeds converged on about 0.2 s per item for several verbal and visual materials. This time was shorter than the speed of articulatory rehearsal measured with the same method, and – in contrast to rehearsal – did not depend on word length. We sought evidence for people refreshing in sync with the metronome by presenting recognition probes at unpredictable times. We expected that probes matching the just-refreshed item should be recognized faster and more accurately than probes matching other items. This was not the case. A parallel experiment with overt articulatory rehearsal showed poor synchronization of rehearsal with the metronome, suggesting by analogy that refreshing was equally out of sync. The results support the assumption that people can attend sequential to items in working memory, and monitor this process. This refreshing process is probably faster than rehearsal, but it is unlikely to be possible as fast as the refreshing process assumed in the TBRS theory.