Influenza viruses continuously circulate in the human population. Acquired immunity through infection or vaccination can be long-lived, yet individuals experience reinfections with antigenically distinct viral strains and subtypes throughout their lives. Furthermore, reinfections can boost antibody responses against viral strains first encountered in childhood, a process termed original antigenic sin (OAS). It is unknown how initial childhood exposures affect the induction of antibodies against the hemagglutinin (HA) stalk domain of influenza viruses. This is an important consideration since broadly reactive HA stalk antibodies can protect against infection from various subtypes, and universal vaccine platforms are being developed to induce these antibodies. Here we show that experimentally infected ferrets and naturally infected humans establish strong “immunological imprints” against HA stalk antigens first encountered during primary influenza virus infections. By defining the antibody specificity through serological assays, we find that HA stalk antibodies are boosted upon subsequent infections with antigenically distinct influenza A virus subtypes. Paradoxically, these heterosubtypic-boosted HA stalk antibodies do not bind efficiently to the boosting influenza virus strain, an indication of OAS. Next, in an effort deliver a better immunological imprint, we design an mRNA vaccine encoding 20 different HAs. We then use mouse and ferret animal models to define the antibody specificity elicited by this vaccine. Our 20 HA mRNA vaccine is able to elicit both strain specific and cross reactive antibodies. Furthermore, it is able to provide in vivo protection from challenge with an influenza virus not directly encoded by the HA sequences included, suggestive of its ability to elicit broad and potent immunity. Overall, the work presented here provides evidence of an immunological bias towards the HA stalk portion of the influenza virus first encountered, but also introduces a solution that may provide a broader immunological imprint.