Alcohol outcome expectancies have been linked to drinking behavior on both empirical and theoretical grounds. Although typically measured as a static construct, we hypothesized that expectancies may be time-specific. Subjects rated their expectancies for a moderate amount of alcohol to increase, decrease, or not change their level of tension and anxiety. Ratings were repeated for when the intoxicating effects of the drinking would be: (1) "at their peak;" (2) "nearly worn off;" and (3) "completely worn off" (Time Epochs 1-3, respectively). As predicted, most subjects (72%) expected alcohol to reduce tension and anxiety at Time Epoch 1; however, significantly fewer subjects expected this effect at Time Epochs 2 and 3 (25% and 2%, respectively). Conversely, few subjects expected alcohol to worsen tension and anxiety at Time Epoch 1 (3.5%); however, significantly more subjects expected this effect at Time Epochs 2 and 3 (31% and 34%, respectively). Expectancies for Time Epoch 1 related most strongly to several measures of alcohol use, including drinking for the purpose of reducing tension (whole sample) and drinking frequency (men but not women). These findings show that tension-reduction expectancies are not stable over the course of a drinking episode and suggest the possibility of a treatment approach aimed at amplifying attention to expectancies for alcohol's more negative longer-term effects.