This study tests the hypothesis that the third molars of Callimico goeldii represent a reversal in evolutionary tooth loss within the Callitrichinae. Loss of third molars is part of a suite of unusual characters that has been used to unite marmosets and tamarins in a clade to the exclusion of Callimico. However, molecular phylogenetic studies provide consistent support for the hypothesis that marmosets are more closely related to Callimico than to tamarins, raising the possibility that some or all of the features shared by marmosets and tamarins are homoplastic. Here, I use the binary-state speciation and extinction (BiSSE) model and a sample of 249 extant primate species to demonstrate that, given the shape of the primate phylogenetic tree and the distribution of character states in extant taxa, models in which M3 loss is constrained to be irreversible are much less likely than models in which reversals are allowed to occur. This result provides support for the idea that the last common ancestor of Callimico and marmosets was characterized by the two-molared phenotype. The M3s of Callimico therefore appear to be secondarily derived rather than plesiomorphic. This conclusion may also apply to the other apparently plesiomorphic traits found in Callimico. Hypotheses regarding the re-evolution of M3 in the callitrichine clade and the origin and maintenance of the two-molared phenotype are discussed.