Most predictions can be partitioned into two components: the predicted outcome, and the chance that one considers the outcome will happen. We studied how people evaluate predictions with binary outcomes. These predictions can be conveyed in two equivalent ways: one predicting an outcome with some probability, and the other predicting the other outcome with the probability of the complement of the first outcome. Although these two ways of stating the predictions are mathematically interchangeable, we hypothesized that people would judge the congruently stated prediction, one that has the same qualitative component as the actual outcome, as more accurate. We tested this hypothesis in four experiments. Results suggested that this effect is consistent across a number of domains; depends on the frame in which the prediction is stated; is robust regardless of whether the ratings were elicited in positive or negative terms; holds for both rating and choice tasks.