Abstract: The liquefaction of cohesionless soils is closely related to the pore water pressure. The in situ testing technology, piezocone penetration test (CPTU) which can directly measure the pore water pressure, has unique advantages in liquefaction evaluation. In this study, a CPTU-based model for assessing the probability of liquefaction is derived by a rigorous mathematical method. The new probabilistic model is expressed in the form of a mapping function that relates the liquefaction probability mathematically to the factor of safety obtained from the CPTU-based deterministic model, which is established by the kernel extreme learning machine algorithm and the robust search technology. The new probabilistic model considers the inherent model uncertainty and parameter uncertainty, and makes a comparative analysis of the probabilistic model. The results show that the factor of safety (F_S) of 1 yields a probability of liquefaction (P_L) of 15.2%. The obvious advantage of the new model is that it directly uses the CPTU data, which is more suitable for the phenomenon of liquefaction behavior, and is suitable for sandy soil and silty soil. The new model can be used for liquefaction evaluation or preliminary screening without the need of additional sampling and laboratory testing. Finally, a liquefaction case of Tangshan earthquake in China is used to illustrate the application of the proposed probabilistic model considering the uncertainties of model and parameters.