<p>Accurate characterization of soil shear modulus and damping ratio at small-to-medium strain levels is essential for reliable prediction of ground response under dynamic loads and for the safe design of geotechnical structures. Despite the development of numerous empirical models, their predictive accuracy remains a critical concern owing to the diversity of soil types and the complexity of field conditions. Recently, data-driven deep neural networks (DNNs) have emerged as a promising approach for modeling complex systems. However, their generalization capability is often limited by the scarcity of high-quality experimental datasets. To address this issue, this study proposes a multifidelity neural network (MFNN) to leverage the accuracy of the high-fidelity experimental datasets and the abundance of the low-fidelity datasets synthesized using the simple empirical model. The MFNN model can automatically learn the correlation between the low-fidelity and high-fidelity datasets. The model has been successfully applied to predict the small-to-medium strain shear modulus and damping ratio of soils. The results demonstrate significantly improved prediction capabilities of MFNN compared to purely data-driven DNNs. The proposed MFNN framework provides a new solution for modeling soil dynamic properties.</p>