We report an experiment examining dual-process models of probability judgments. Participants had to judge which of two samples provided the best chances of picking a target. Often, participants based their judgment on a comparison of the absolute frequencies of the targets instead of on the ratios. The comparison of the absolute frequencies of the targets was either congruent with the comparison of the ratios (congruent condition), or not (incongruent condition). In this study, the proportion of incongruent (versus congruent) trials was manipulated. Accuracy rates and reaction times were recorded. Accuracy rates for incongruent items decreased but remained unchanged for congruent items when more congruent items were presented. The most striking reaction time result is that the RTs for correct responses to congruent items drastically increased when fewer such items were presented, whereas the RTs for correct and incorrect responses to incongruent items remained unchanged. This increase in RTs for correct responses to congruent items runs counter to the parallel-competitive model, but can in our opinion be reconciled with a default-interventionist model.