We investigated possible explanations of the finding that the relative weight (W) of common components in similarity judgments is higher for verbal than for pictorial stimuli. A serial presentation of stimulus components had no effect on verbal stimuli; it increased the impact of both common and distinctive components of pictorial stimuli but did not affect their relative weight. On the other hand, W was increased by manipulations that reduced the cohesiveness of composite pictures, such as separating, scrambling, and mixing their components. Furthermore, W was decreased by manipulations that enhanced the cohesiveness of composite verbal stimuli by imposing structure on their components. Verbal and pictorial representations of the same stimuli yielded no systematic differences in W.