The redshift distribution of near-IR-selected galaxies is often used to attempt to discriminate between the classical view of galaxy formation, in which present-day luminous galaxies were assembled at early times and evolve through the passive aging of their stellar populations, and that of hierarchical structure formation, in which galaxies were assembled more recently via the merging of smaller objects. We carry out such a test here by computing the distribution of photometric redshifts of KAB < 22 galaxies in the Great Observatories Origins Deep Field Survey (GOODS) southern field, and comparing the results with predictions from a semianalytic model based on hierarchical structure formation, and a classical "passive evolution" model. We find that the redshift distributions at z 1.5 of both the hierarchical and passive models are very similar to the observed one. At z 1.5 the hierarchical model shows a deficit of galaxies, while the passive model predicts an excess. We investigate the nature of the observed galaxies in the redshift range where the models diverge, and find that the majority have highly disturbed morphologies, suggesting that they may be merger-induced starbursts. While the hierarchical model used here does not produce these objects in great enough numbers, the appearance of this population is clearly in better qualitative agreement with the hierarchical picture than with the classical passive evolution scenario. We conclude that the observations support the general framework of hierarchical formation, but suggest the need for new or modified physics in the models.