Abstract: Structural dynamics aims to analyze the structural responses under dynamic loads such as earthquakes, and it requires solving governing differential equations of motions. Physics-informed neural networks combine physical constraints with data-driven learning for this purpose, but spectral bias limits their ability to represent high-frequency seismic components. An eigenfrequency-embedded physics-informed neural network is proposed. Structural natural frequencies are embedded into activation functions to construct multifrequency basis functions. This design enhances high-frequency representation. Neural tangent kernel analysis shows slower eigenvalue decay and improved gradient propagation in high-frequency ranges. The high-frequency learning limitation of standard physics-informed neural networks is therefore alleviated. Numerical examples show improved accuracy in displacement, velocity, and acceleration responses compared with conventional physics-informed neural networks. Stable performance is observed under different seismic excitations, moderate noise levels, and partial observation losses. These results indicate the applicability of the proposed method to structural dynamic response analysis.