Neuraminidase (NA) is increasingly recognized as a promising target antigen for improving the breadth and durability of influenza vaccines, yet NA-based mRNA vaccines remain comparatively unexplored relative to hemagglutinin-based approaches. Here, we applied a computationally guided NA antigen-design strategy to seasonal H3N2 influenza viruses and investigated whether epitope-guided optimization could improve the immunogenicity and protective efficacy of NA-based mRNA vaccines. NA-D3, an NA antigen designed based on the evolutionary characteristics of circulating H3N2 viruses, was first evaluated in an mRNA vaccine. This vaccine provided partial protection against H3N2 challenge and reduced viral replication in the lungs, although detectable neuraminidase inhibition (NAI) activity was not observed. To further improve NA antigen immunogenicity, predicted B- and T-cell epitopes were incorporated into the antigen design process, resulting in three sequentially optimized NA candidates, F0, F1, and F2. In a subsequent challenge experiment, F1 and F2 conferred complete protection, achieving 100% survival and significantly lower viral RNA levels and infectious viral titers in the lungs compared to controls. The optimized NA mRNA vaccines also induced detectable NAI activity against heterologous H3N2 viruses, albeit at relatively modest levels. Collectively, these findings demonstrate the protective potential of NA-based mRNA vaccination against H3N2 influenza viruses and establish iterative, epitope-guided strategy in NA antigen optimization – integrating computational prediction with experimental validation – as an effective approach for developing broadly protective NA-based mRNA vaccines.