Abstract: The liquefaction models established through the in-situ tests and liquefaction case histories are widely adopted in the liquefaction evaluation of a site. However, it is reported that the model accuracy varies with the in-situ test methods (i.e., model bias), as well as the earthquake sites (i.e., seismic site variability). To assess the model bias of SPT-Youd model, V_s-AS model and CPT-RW model, a database of liquefaction case histories containing all these three in-situ tests is compiled firstly, containing 88, 176 and 107 cases, respectively, and involving sites from six seismic events. Based on the compiled database, the Bayesian hierarchical modeling (BHM) method is adopted to correct the model bias of the three models, and liquefaction models for each of the seismic sites are established. It is found that the CPT-RW model is the most conservative model, and the liquefaction resistance calculated by the SPT-Youd model is overestimated in several seismic sites, and the conservatism of the V_s-AS model is between them according to the complied database. Using the complied CPT database, the seismic site-specific liquefaction models established by the BHM method can effectively correct the conservatism of the original CPT-RW model, and the levels of correction vary from one seismic site to another. This study provides effective priori information for engineers to assess the liquefaction potential of a site using multiple in-situ tests.