When the same syllable is presented repeatedly to a human subject, it undergoes perceptual changes known as verbal transformations. However, the dynamics of such perceptual changes is poorly understood. In the present work we show that the main organization of the perceptual transitions is into pairs. This pairwise coupling is pronounced in the frequency of switching to pair members, but not in their dwell times (the time spent perceiving a given phonemic form before switching to another form). We show that the paircoupled transforms of perception have a faster and more stable dynamic than the nonpaircoupled transforms. We also demonstrate that the pairwise coupling is stronger than would be expected from random arrangement of small numbers of transforms. These characteristic patterns of verbal transforms have been predicted by a mathematical model, first proposed as a model of perceptual alternations of ambiguous visual figures.