Safety evaluation of a human vaccine is critical for vaccine development and for preventing an unexpected adverse reaction in humans. Nonetheless, to date, very few systems have been described for preclinical studies of human adverse reactions in vivo. Previously, we have identified biomarker genes expressed in the lungs for evaluation of influenza vaccine safety, and their usefulness in rodent models and for adjuvant-containing vaccines has already been reported. Here, our purpose was to develop a novel humanized mouse model retaining human innate-immunity-related cells to assess the safety of influenza vaccines using the previously identified biomarker genes. In the present study, we tested whether the two humanized models, a short-term and long-term reconstitution model of NOD/Shi-scid IL2r?null mice, are suitable for biomarker gene-based safety evaluation. In the short-term model, human CD14+ cells, plasmacytoid dendritic cells, CD4+ and CD8+ T cells, and B cells were retained in the lungs. Among these cells, human CD14+ cells and plasmacytoid dendritic cells were not detected in the lungs of the long-term model. After the vaccination, the expression levels of human biomarker genes were elevated only in the short-term model when the toxicity reference vaccine was inoculated. This phenomenon was not observed in the long-term model. The levels of human cytokines and chemokines in the lungs increased in response to the toxicity reference vaccine in the short-term mouse model. According to these results, the short-term model provides a better platform for evaluating vaccine safety in terms of human peripheral blood mononuclear cell-mediated initial reactions in vivo