Vibrotactile discrimination tasks involve perceptual judgements on stimulus pairs separated by a brief interstimulus interval (ISI). Despite their apparent simplicity, decision making during these tasks is biased by prior experience in a manner that is not well understood. A striking example is when participants perform well on trials where the first stimulus is closer to the mean of the stimulus-set than the second stimulus, and perform comparatively poorly when the first stimulus is further from the stimulus mean. This “time-order effect” suggests that participants implicitly encode the mean of the stimulus-set and use this internal standard to bias decisions on any given trial. For relatively short ISIs, the magnitude of the time-order effect typically increases with the distance of the first stimulus from the global mean. Working from the premise that the time-order effect reflects the loss of precision in working memory representations, we predicted that the influence of the time-order effect, and this superimposed “distance” effect, would monotonically increase for trials with longer ISIs. However, by varying the ISI across four intervals (300, 600, 1200, and 2400 ms) we instead found a complex, non-linear dependence of the time-order effect on both the ISI and the distance, with the time-order effect being paradoxically stronger at short ISIs. We also found that this relationship depended strongly on participants' prior experience of the ISI (from previous task titration). The time-order effect not only depends on participants' expectations concerning the distribution of stimuli, but also on the expected timing of the trials.