Maternal care is essential for infant survival across mammalian species. In rodents, a classical maternal behavior, retrieving astray pups back to the nest (“pup retrieval”), is broadly expressed by mothers, even those with no previous experience (e.g., primiparous mothers), suggesting that this behavior emerges through hard-wired neural mechanisms that are activated during motherhood. However, sexually- and pup-naïve female mice (virgins) are also able to perform this behavior if exposed to mothers and their litter, insinuating that there is a learning component to this behavior that is independent of socio-sexual history. To understand the complex interplay of innate vs. learned elements that lead to the emergence of maternal behaviors, we first hypothesize that the behavior can be decomposed into sub-components (e.g., picking up a pup, dropping it on the nest) and that each is distinctly represented in neural activity. Second, we hypothesize that learning of the complete behavior (e.g. the retrieval) happens when the sub-components are temporally and spatially correctly linked. To test this, we recorded mouse trajectories during pup retrieval of mothers and virgins. We developed a new methodology based on Kalman filters to compare the variability/predictability of the behavioral trajectories and their sub-components. We show that virgins initially engage in pup retrieval very stochastically, then progressively improve until they mirror the behavioral dynamics of mothers. Furthermore, preliminary microendoscopic calcium imaging data shows how this innate behavior can also be learned with repeated exposure to pups, and the differences in representations between the two groups.