In a multiple criteria decision-making problem, decision-makers often make judgments of importance, for example, that “rent is more important than apartment size” when choosing apartments. Even though linear models are heavily used in choice prediction, it has remained unclear whether criterion weights are connected to judgments of importance. A surprisingly common assumption is that a more important criterion tends to have a larger weight, as if weights and importance were equal, or at least heavily correlated. In the experiment, subjects provided pairwise judgments of importance for four criteria and made pairwise choices with apartments defined by these criteria. According to our results, Goldstein’s (1990) idea of connecting judgments of importance to impact is more meaningful than connecting them to weights. Impact as the product of AHP weights and coefficient of variation is the best definition for impact, when measured by correlation to the original judgments of importance.