We present progress towards a novel theoretical approach for understanding Tversky’s famous ‘diagnosticity’ effect in similarity judgments, and an initial empirical validation. Our approach uses a model for similarity judgments based on quantum probability theory. The model predicts a diagnosticity effect under certain conditions only. Our model also predicts that changes to the set of stimuli to be compared can cause the diagnosticity effect to break down or reverse. In one experiment, we test and confirm one of our key predictions