As recognized universally by both seismology and earthquake engineering communities, the amplitude and frequency content of ground motions are influenced by local site effects, including the effects of near-surface geologic materials, surface topographic and basin effects, and so on. Strong linkage between seismic site effect and earthquake damage has been commonly demonstrated from many past earthquakes. Therefore, quantitative and reliable evaluation of the seismic site effect is one of the crucial aspects in seismic hazard assessment and risk mitigation. With the significant advancement of modern seismic monitoring networks and arrays, huge amounts of high-quality seismic records are now being accumulated. This encourages us to measure the site responses and its associated uncertainty for selected seismic stations by some record-dependent approaches, such as horizontal-to-vertical spectral ratio (HVSR) measurements, generalized spectral inversion (GIT) methods, etc. Machine learning techniques also show significant promise in characterization of the near-surface geologic properties and prediction of site response. These data-driven approaches help us to better understand the physics of spatial and temporal variabilities of ground motions. Due to more and more site-specific data being captured, invoking non-ergodic assumptions in seismic response analysis has recently been a topic of great interest in the community. For specific site response analysis, numerical simulations are carried out to model the dynamic process of seismic waves propagating and scattering in the subsurface strata. With development of modeling capacity, great efforts have been taken to evaluate quantitatively the complex 2D and 3D effects on seismic site response.