Abstract: An inversion algorithm is proposed to estimate the model parameters of soil dynamic response based on the KiK-net downhole array recordings. The algorithm borrows the Bayesian estimation technique using the unscented Kalman filtering, and can make full use of the prior message to obtain the estimation of the model parameters and their uncertainties. By solving a numerical example, it is shown that the algorithm is successful in estimating the dynamic soil properties. After applying the algorithm to a KiK-net station, the results show that the estimation of the shear wave velocity profile can well characterize the linear behavior of soils. Using the estimation of the model parameters of soils, the dynamic response model shows good performance of predicting the response of soil surface.