The aim of the studies presented in this paper was to propose a new explanation of under- and overestimation effects in comparative judgments. The fundamental assumption of this new interpretation is that in comparative judgments ('the object X in comparison with the object Y' type) the target is contrasted with the comparison standard when the compared objects seem generally dissimilar and assimilated to the standard when the objects seem generally similar. In a series of three studies students were asked to compare the chances of certain events occurring to two objects (self vs. classmate vs. the average person). The direction of comparison was manipulated. Generally, when the more salient object was compared to the less salient object, irrespective of the valence of the events, the overestimation effects occurred only in case of frequent events and the underestimation effects only in case of rare events. The reversal of direction of comparisons yielded the clear reduction of comparative bias.