Interpretations of dynamic processes and the thermal and chemical structure of the Earth depend on the accuracy of Earth models. With the growing number of velocity models constructed with different tomographic methods and seismic data sets, there is an increasing need for a systematic way to validate model accuracy and resolution. This study selects five shear wave velocity models in the U.S. and simulates full wave propagation within the 3-D structures. Surface-wave signals extracted from ambient seismic noise and regional earthquakes are compared with synthetic waveforms at multiple-frequency bands. Phase delays and cross-correlation coefficients between observed and synthetic waveforms allow us to compare and validate these models quantitatively. In general, measurements from regional earthquakes are consistent with ambient noise results, but appear more scattered, which may result from uncertainty of the earthquake source location, origin time, and moment tensor. Our results show the improvement of model prediction with the increase of seismic data sets and implement of advanced methods. There exists a positive linear trend between phase delay and interstation distance for three models, indicating that on average, these models are faster than the real Earth structure. The phase delays from the jointly inverted model of ambient noise and receiver function have negative means at all periods while without obvious dependence on the interstation distance. The full-wave ambient noise tomographic model predicts more accurate phase arrivals compared to other models. This study suggests a need for an integrated model constructed with multiple seismic waveforms and consideration of anisotropy and attenuation.