Tooth number in rodents is an example of reduction in evolution. All rodents have a toothless diastema lacking canine and most premolars present in most other mammals. Whereas some rodent lineages retained one premolar (p4), many others lost it during evolution. Recently, an 'inhibitory cascade' developmental model (IC) has been used to predict how the first molar (m1) influences the number and relative sizes of the following distal molars (m2 and m3). The model does not, however, consider the presence of premolars, and here we examine whether the premolar could influence and constrain molar proportions during development and evolution. By investigating a large data set of both extinct and extant rodent families over more than 40 million years, we show that the basal phenotype is characterized by the presence of premolars together with equally sized molars. More recent rodent families, with and without premolar, show more unequal molar sizes. Analysing molar areas, we demonstrated that (i) rodents harbour almost all the molar proportions known in mammals, and the IC model can explain about 80% of taxa in our data set; (ii) proportions of molars are influenced by the presence or absence of p4; and (iii) the most variable teeth in the dental row are m1 and m3, whether p4 is present or not. Moreover, m1 can represent up to half of the total molar area when p4 is absent. We hypothesize that p4 loss during evolution released the constraint on m1 development, resulting in a more variable size of m1 and thereby having an indirect effect on the evolution of the whole molar row.