Abstract: Due to the differences in research methods and data, there may exist multiple velocity models in the same area, but the reliability of these models usually lacks systematic and objective assessment. In this study we compare the observed first arrival time data of P-wave and S-wave of 1 749 earthquakes from 2009 to 2016, which were recorded by 131 seismograph stations of the National Seismological Network in North China, with the predicted travel time data from four 3D crustal velocity models in North China using the fast marching method. Then, using statistical analysis we evaluate the relative merits of these four models with respect to the real underground structures. The results reveal that the large-scale pattern generally shows consistency for these four models. Within the entire studied area, the model proposed by Shen et al （referred to as " S model”） is relatively better than the model proposed by Fang et al （referred to as " F model”） and Duan et al （referred to as " D model”）. The Crust1.0 model （referred to as " C model”） is relatively worse. We think that the reasons for this result are related to the differences in data used in model construction and the associated resolution. For different tectonic units in the studied area, the D model performs better in the southwestern part of the Yanshan folded belt, northwestern part of the Taihang mountain foreland tectonic belt, and the Cangxian uplift area. The F model performs better in the central part of Taihang mountain uplift, northern part of the Cangxian uplifted area, the Huanghua depression, and the Yanshan fold belt. The S model appears better in the western block, the Shanxi depression area, the Taihang mountain foreland tectonic belt, and the Jizhong depression area. There is no obvious large continuous area where the C model performs better. Our study has certain positive significance for further accuracy improvement of the regional crustal velocity models and the earthquake location study based on 3D models.