Affective information in our environment is often predictable by time; for example, positive answers are typically given faster than negative ones. Here we demonstrate, for the first time, that humans can implicitly adapt to time-based affect predictability. Participants were asked to categorize words, with the words’ irrelevant valences being predictable by the timing of their occurrence. Adaptation to this pattern became evident by better performance for typical combinations of time and valence, relative to atypical combinations (Experiment 2). A comparable adaptation was observed for predictable activation (another affective dimension, Experiment 4), but not for predictable imageability (a nonaffective dimension, Experiment 3). In none of the experiments did participants become aware of the time-based predictability. These findings have significant implications for our theoretical understanding of human time-based expectancy, as well as important implications for the scheduling of system delays in artificial interaction and communication environments.