When the duration of a pre-target interval probabilistically predicts the identity of the target, participants typically form time-based expectancies: they respond faster to frequent interval-target combinations than to infrequent ones. Yet, previous research investigating the cognitive time-processing mechanisms underlying time-based expectancy assessed time-based expectancy always in situations with a binary set of intervals (i.e. short vs. long). Here we aim to test whether time-based expectancy transfers to more complex settings with three different predictive time intervals (short, medium, long) in which each predicts one of three different target stimuli with 80% probability. In three experiments we varied how the medium interval was computed (arithmetic mean, geometric mean, or in between both). Our results showed that participants were able to learn the time-event contingencies for the short and the long as well as for the medium interval, and were, thus able to flexibly redirect their target expectancy two times during the course of a trial. The evidence concerning the impact of the manipulation of the medium intervals’ absolute duration on time-based expectancy was, however, mixed, as time-based expectancy for the medium interval could only be observed in one of three reported experiments. In sum, the findings of the present study suggest a previously unknown cognitive flexibility underlying time-based expectancy and offer important theoretical implications, challenging future research on the timing mechanisms involved in time-based expectancy.