Human children copy others' actions with high fidelity, supporting early cultural learning and assisting in the development and maintenance of behavioral traditions [1]. Imitation has long been assumed to occur from birth [2-4], with influential theories (e.g., [5-7]) placing an innate imitation module at the foundation of social cognition (potentially underpinned by a mirror neuron system [8, 9]). Yet, the very phenomenon of neonatal imitation has remained controversial. Empirical support is mixed and interpretations are varied [10-16], potentially because previous investigations have relied heavily on cross-sectional designs with relatively small samples and with limited controls [17, 18]. Here, we report surprising results from the most comprehensive longitudinal study of neonatal imitation to date. We presented infants (n = 106) with nine social and two non-social models and scored their responses at 1, 3, 6, and 9 weeks of age. Longitudinal analyses indicated that the infants did not imitate any of the models, as they were just as likely to produce the gestures in response to control models as they were to matching models. Previous positive findings were replicated in limited cross-sections of the data, but the overall analyses confirmed these findings to be mere artifacts of restricted comparison conditions. Our results undermine the idea of an innate imitation module and suggest that earlier studies reporting neonatal imitation were methodologically limited.